Intelligent assistant creating method and system and storage medium

Abstract

The invention discloses an intelligent assistant creating method and system and a storage medium, and aims to efficiently and conveniently create an intelligent assistant meeting user requirements in a full-process automation mode and adapt to diversification and dynamic evolution of the user requirements. The intelligent assistant creation method comprises the following steps: in response to a first task demand of creating an intelligent assistant indicated by a user, obtaining a first process template matched with the first task demand from a template library, and generating a first process executed by the created intelligent assistant based on the first process template; based on the description information of each agent component in the agent component library, matching at least one first agent component for a first subtask in the first process; and assembling the at least one first agent component corresponding to the first sub-task to obtain a first agent corresponding to the first sub-task, and creating an interaction logic between the first agents corresponding to the first sub-task based on a logical relationship between the first sub-tasks to obtain a first intelligent assistant.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, system and storage medium for creating an intelligent assistant. Background Technology

[0002] With the rapid development of artificial intelligence technology, creating intelligent assistants based on large model technology to assist business processing has become an important way to significantly improve business efficiency.

[0003] However, given the diversification and dynamic evolution of user needs, a comprehensive solution is still needed to quickly and conveniently create intelligent assistants that meet those needs. Summary of the Invention

[0004] The purpose of this application is to provide a method, system, and storage medium for creating a smart assistant, which uses a fully automated process to efficiently and conveniently create a smart assistant that meets user needs and can adapt to the diversification and dynamic evolution of user needs.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: In a first aspect, embodiments of this application provide a method for creating a smart assistant, including: In response to a user's instruction to create a smart assistant, a first process template matching the first task requirement is retrieved from a template library, and a first process executed by the created smart assistant is generated based on the first process template; the first process includes multiple first sub-tasks and the logical relationships between the multiple first sub-tasks. Based on the description information of each intelligent agent component in the intelligent agent component library, at least one first intelligent agent component is matched for the first subtask in the first process; Assemble at least one first intelligent agent component corresponding to the first subtask in the first process to obtain the first intelligent agent corresponding to the first subtask in the first process, and based on the logical relationship, create the interaction logic between the first intelligent agents corresponding to the first subtask in the first process to obtain the first intelligent assistant.

[0006] Secondly, embodiments of this application provide an intelligent assistant creation system, including: The system includes a demand processing intelligent agent subsystem, a reasoning and planning intelligent agent subsystem, an assembly and coordination subsystem, and a data pool, wherein the data pool contains a template library and an intelligent agent component library. The demand processing intelligent agent subsystem is used to obtain the first task demand for creating an intelligent assistant from the user's instructions. The reasoning and planning intelligent agent subsystem is used to respond to the first task requirement, obtain a first process template matching the first task requirement from the template library, and generate a first process executed by the created intelligent assistant based on the first process template; the first process includes multiple first sub-tasks and the logical relationship between the multiple first sub-tasks. The assembly and coordination subsystem is used to match at least one first intelligent agent component for the first subtask in the first process based on the description information of each intelligent agent component in the intelligent agent component library, and to assemble at least one first intelligent agent component corresponding to the first subtask in the first process to obtain the first intelligent agent corresponding to the first subtask in the first process. Based on the logical relationship, it creates the interaction logic between the first intelligent agents corresponding to the first subtask in the first process to obtain the first intelligent assistant.

[0007] Thirdly, embodiments of this application provide a computer-readable storage medium that, when the instructions in the storage medium are executed by a processor of an electronic device, enables the electronic device to perform the intelligent assistant creation method provided in the first aspect.

[0008] The above-described technical solutions adopted in the embodiments of this application can achieve the following beneficial effects: By integrating a large number of process templates into a process template library, and retrieving a matching first process template from the library based on task requirements during the creation process, the first process executed by the created intelligent assistant can be generated. This reduces the randomness in process generation and ensures that the generated first process fully meets the task requirements. Furthermore, by integrating a large number of intelligent agent components into an intelligent agent component library, and matching at least one first intelligent agent component to the first subtask in the first process based on the description information of each intelligent agent component, the first intelligent agent corresponding to the first subtask is obtained by assembling these first intelligent agent components. Based on the logical relationships between the first subtasks, the interaction logic between the first intelligent agents corresponding to the first subtasks is created, thus quickly obtaining a first intelligent agent assistant that fully meets the requirements of the first task. It is evident that users only need to input the task requirements for creating an intelligent assistant to trigger a fully automated intelligent assistant creation process, significantly lowering the barrier to entry for intelligent assistant creation and reducing the requirement for users' professional knowledge, enabling even non-professional users to easily complete the creation of intelligent assistants. Secondly, the fully automated approach efficiently and conveniently creates intelligent assistants that meet user needs; the intelligent assistant can be rebuilt simply by modifying the task requirements, thus better adapting to the diversification and dynamic evolution of user needs. Attached Figure Description

[0009] 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: Figure 1 A schematic diagram of an example environment in which embodiments of this application can be implemented is shown; Figure 2 A flowchart illustrating a method for creating a smart assistant, as provided in an embodiment of this application; Figure 3 This application provides a schematic diagram of the structure of an intelligent assistant creation system. Figure 4 A schematic diagram illustrating the generation process of a prompt word template provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a data pool provided in an embodiment of this application; Figure 6 This is a schematic diagram illustrating the interaction between a subsystem and a data pool in an intelligent assistant creation system provided in an embodiment of this application. Detailed Implementation

[0010] 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.

[0011] The term "comprising" and its variations as used in this document are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. The term "in response to" indicates that the performed operation depends on a condition or state. When the dependent condition or state is met, one or more operations may be performed in real time or with a set delay. Unless otherwise specified, there is no restriction on the order in which multiple operations are performed.

[0012] It should be noted that the concepts of "first" and "second" mentioned in this document are used only to distinguish different devices, modules or units, and are not used to restrict the order of functions performed by these devices, modules or units or their interdependencies.

[0013] It should be noted that the terms "one" and "more" used in this document are illustrative rather than restrictive, and those skilled in the art should understand that, unless explicitly stated otherwise in the context, they should be understood as "one or more".

[0014] The names of messages or information exchanged between multiple devices in the embodiments of this document are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0015] The technical solutions of the various embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0016] Figure 1 A schematic diagram of an example environment in which embodiments of this application can be implemented is shown. This example environment includes a terminal device 100 and a server 200.

[0017] The terminal device 100 and the server 200 establish a communication connection, which may include, but is not limited to, at least one of the following: wired connection and wireless connection.

[0018] Terminal device 100 includes, but is not limited to, smartphones, tablets, laptops, desktop computers, smart voice interaction devices, smart home appliances, smartwatches, vehicle terminals, and aircraft. Server 200 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0019] Terminal device 100 can receive relevant parameters input by the user and send these parameters to server 200. Server 200 can obtain the processing result based on the received parameters and return the processing result to terminal device 100.

[0020] It should be understood that in some embodiments, the terminal device 100 may obtain the processing result itself based on the relevant parameters input by the user, without needing to interact with the server 200. This application embodiment does not limit this.

[0021] Based on the example environment described above, this application provides a method for creating a smart assistant. Please refer to... Figure 2 The above is a flowchart illustrating a method for creating a smart assistant according to an embodiment of this application. The method includes the following steps: S202, in response to the user's instruction to create a first task requirement for a smart assistant, a first process template matching the first task requirement is obtained from the template library, and a first process is generated based on the first process template to be executed by the created smart assistant.

[0022] The first task requirement can be understood as the requirements that the created intelligent assistant needs to meet, such as the functions and performance (such as accuracy threshold, functional flexibility preference, and expected construction speed) that the created intelligent assistant is expected to have.

[0023] The requirements for the first task can be obtained in various ways, and this application does not limit them in the embodiments.

[0024] In one implementation, the first task requirement is obtained by: receiving dialogue data input by the user in the dialogue interface; performing intent recognition on the dialogue data to obtain a first intent; and generating the first task requirement based on the first intent in response to the first intent to create a smart assistant.

[0025] Terminal device 100 provides a dialog interface through which users input dialogue data in one or more forms, such as voice, text, and images. When users input dialogue data in voice form, terminal device 100 can use Automatic Speech Recognition (ASR) technology to convert the dialogue data into text form and then perform intent recognition; when users input dialogue data in image form, terminal device 100 can use image recognition technology to convert the dialogue data into text form and then perform intent recognition.

[0026] The dialogue data in this application embodiment may include the dialogue data of the user in the current round, or the dialogue data of the user in the previous round or multiple rounds.

[0027] The intent recognition of dialogue data can be achieved using various intent recognition techniques in this field, which will not be elaborated here.

[0028] After identifying the first intent, if the first intent meets preset conditions, it can be used as the first task requirement. The preset conditions can be set according to actual needs. For example, the first intent may specify the requirements that the smart assistant to be created needs to meet, or the first intent may contain the necessary information required to create the smart assistant (such as the functions and performance of the smart assistant to be created). This application embodiment does not limit this.

[0029] If the initial intent does not meet the preset conditions, it is expanded to obtain the initial task requirements. This ensures the completeness and accuracy of the initial task requirements, providing comprehensive data support for the subsequent precise creation of the intelligent assistant.

[0030] Specifically, in response to the dialogue data not containing the user's first user feature, the user's second user feature is obtained from the user feature library, and the first intent is expanded based on the second user feature to obtain the first task requirement; in response to the dialogue data containing the user's first user feature, the first intent is expanded based on the first user feature to obtain the first task requirement, and the first user feature is written into the user feature library to update the second user feature.

[0031] The first user feature can be understood as the user characteristics extracted from the dialogue data that reflect the user's personalized features. The user feature library includes multiple user features, which can come from different users. The second user feature can be understood as the user features of the user to whom the dialogue data belongs in the user feature library.

[0032] In this embodiment, user features may include, but are not limited to, at least one of the following: structured information (such as identity tags and domain affiliation) and unstructured information (such as operating habits and performance preferences). For structured information, entity recognition algorithms based on pre-trained models such as BERT (Bidirectional Encoder Representation from Transformers) can be used to parse key information (such as keywords) in the dialogue data, and a classification model can be combined to determine the user's identity tags, domain affiliation, etc. For unstructured information, natural language understanding techniques (such as intent classification models and sentiment classification algorithms) can be used to parse the style and emotional tendencies of the dialogue data, and behavioral sequence analysis algorithms can be combined to capture patterns such as interaction frequency and operation paths, and extract unstructured information such as the user's operating habits and performance preferences.

[0033] If the first user feature is extracted from the dialogue data, it can be combined with the first intent to form the first task requirement. If the first user feature cannot be extracted from the dialogue data, the user's second user feature can be queried from the user feature database. This second user feature can then be combined with the first intent to form the first task requirement. Furthermore, writing the first user feature into the user feature database enables dynamic updates to the database. This allows the second user features in the database to provide accurate references for subsequent agent creation, helping to more accurately capture the user's potential intent and ensuring that the final intelligent assistant highly matches the user's actual needs.

[0034] In practice, the user feature database can employ a hybrid storage architecture. For example, it can include a relational database and a document database, with the relational database storing structured information and the document database storing unstructured information. Furthermore, the user's secondary feature information in the database is unique to that user to protect user privacy.

[0035] The template library includes multiple process templates, each containing multiple subtasks and the logical relationships between them. These logical relationships can include, but are not limited to, at least one of the following: execution order (e.g., serial, parallel), dependency relationships (e.g., a subtask depends on the execution result of another subtask), etc. Based on their source and access permissions, the process templates in the template library can be divided into redundant integrated templates and user history templates. Redundant integrated templates are system-provided and shared by all users. User history templates are unique to their respective users and represent the process templates used by that user.

[0036] The process template can be a Standard Operating Procedure (SOP). In some examples, the process template may include multiple nodes and connections between them. Each node represents a subtask, and the connections between nodes represent the logical relationships between subtasks. The nodes in the process template must include at least the following four types: Agent node: It represents a subtask that can be completed by an agent. The node information of this node may include the task description of the corresponding sub-node and the description information of the agent that can complete the subtask (such as the agent components that make up the agent).

[0037] Logical nodes: These represent conditional branches, allowing a process to be split into multiple branches based on predefined conditions. Logical nodes can be considered a special case of agent nodes.

[0038] Input node: The subtask it represents is used to receive input data.

[0039] Output node: The subtask it represents is used to output the result.

[0040] It is worth noting that the subtasks corresponding to agent nodes and logic nodes can be executed by at least one agent component. Here, at least one agent component includes a low-level model component as the core computing unit, while local knowledge base components and tool components are not required.

[0041] The first process template can be understood as a process template in the template library that can meet the requirements of the first task. For example, if the first task requires the functions of the created smart assistant, the sub-tasks included in the first process template can achieve those functions. The first process template can be a redundant integrated template or a user-owned historical template.

[0042] In S202 above, the first process template can be obtained from the template library through various methods such as Retrieval-Augmented Generation (RAG), and this application embodiment does not limit this.

[0043] In one implementation, the first task requirement is decomposed into multiple second sub-tasks; the multiple second sub-tasks are matched with the sub-tasks in each process template in the template library to obtain the matching degree between the first task requirement and each process template; based on the matching degree between the first task requirement and each process template, the first process template in the template library that matches the first task requirement is determined.

[0044] Specifically, the language understanding capabilities of the large model can be utilized to instruct the large model to understand and decompose the requirements of the first task, resulting in multiple second sub-tasks; then, RAG technology can be used to retrieve the first process template most relevant to the multiple second sub-tasks from the template library.

[0045] For example, the first task requirement is "to create a smart assistant with travel planning functions". This first task requirement can be broken down into several second sub-tasks to be executed in the following order: obtain the user's travel needs, such as destination, travel time, travel preferences, etc. → check the weather at the destination → recommend attractions based on the weather at the destination and travel preferences → plan a reasonable itinerary route → recommend restaurants near the itinerary → generate a complete itinerary plan.

[0046] Furthermore, based on the names, descriptions, expected inputs / outputs, etc. of these second subtasks, corresponding first natural language descriptions are generated, and the logical relationships between these second subtasks are encoded into a first vector using graph neural networks or specialized structural embedding techniques. In addition, for each process template in the template library, based on the names, descriptions, expected inputs / outputs, etc. of the subtasks in the process template, a corresponding second natural language description is generated, and the logical relationships between the subtasks in the process template are encoded into a corresponding second vector using graph neural networks or specialized structural embedding techniques.

[0047] Furthermore, the first natural language description is used as the query vector to quickly retrieve the top k most semantically relevant second process templates from the template library, thereby filtering out obviously irrelevant process templates. Then, based on the matching degree between the first vector and the second vectors corresponding to these second process templates, and the coverage of the second process templates to the multiple second sub-tasks, the matching degree between the first task requirement and these second process templates is calculated. Finally, the second process template with the highest matching degree with the first task requirement is taken as the first process template.

[0048] This implementation method allows for a more refined capture of users' real needs, ensuring that the obtained first-process template fully matches these needs, thereby ensuring that the created intelligent assistant can better meet users' real requirements.

[0049] In another implementation, it is not necessary to split the first task requirement. Instead, based on the understanding of the first task requirement, RAG technology is used to retrieve the first process template most relevant to the first task requirement from the template library.

[0050] The first process includes multiple first subtasks and the logical relationships between them. In S202 above, after obtaining the first process template, the first process can be generated in various ways, which is not limited in this embodiment.

[0051] In one implementation, the first process template can be used as the first process.

[0052] In another implementation, the subtasks and their logical relationships in the first process template can be modified according to the requirements of the first task. For example, subtasks that are not related to the requirements of the first task can be deleted, and the first process can be generated quickly.

[0053] It is evident that, compared to generating the first process from scratch based on user intent, matching the optimal process template in the template library and modifying it can significantly improve the accuracy of process generation while reducing the resource overhead required to create a smart assistant.

[0054] S204, based on the description information of each intelligent agent component in the intelligent agent component library, match at least one first intelligent agent component for the first subtask in the first process.

[0055] The agent component library contains multiple agent components. These are pre-built, reusable agent components. Based on their functionality, these agent components can be divided into three categories: underlying model components, local knowledge base components, and tool components. The underlying model components aim to provide agents with large-scale underlying model capabilities, including large language models, multimodal models, coding models, and mathematical models—large models with outstanding capabilities in different domains. The local knowledge base components provide agents with domain-specific external knowledge bases to supplement knowledge gaps and reduce illusions; their specific forms can include vector databases and graph knowledge bases. The tool components aim to provide agents with rich extended functions to enhance their ability to perform specific sub-tasks, such as code interpreters for complex programming tasks and search engines for users to obtain information on the Internet.

[0056] Each agent component has corresponding descriptive information. This descriptive information may include, but is not limited to, at least one of the following: the agent component's identifier, type, and functional information. This descriptive information can be stored in the agent component library in the form of a triple <identifier, category, functional information>. The agent component's identifier is a dedicated index assigned to it for easy retrieval. The functional information details the agent component's characteristics, its specific role in the agent, and possible application scenarios. This functional information can be represented using a JavaScript Object Notation (JSON) data structure. For example, a large model component used for generating images from text, categorized as "underlying model," would have the following functional information: { "function_calling": "False", "function_description": "Generates high-quality, realistic images based on the input text description. Supports image generation in various styles and themes, such as landscapes, people, and animals, and can perform style transfer and image editing according to user instructions." "application": "Can be used for image generation tasks, such as providing users with visual content creation services based on text descriptions, or generating corresponding environmental elements based on descriptions in virtual scene construction." } The intelligent agent component library grants users autonomy, allowing them to easily manage intelligent agent components such as adding, deleting, modifying, and querying them. It's important to note that when adding a personalized intelligent agent component, users need to write complete functional information; and usability testing should be performed on any modified intelligent agent components. Usability testing refers to the intelligent agent component's ability to respond successfully and execute the specified subtasks as expected.

[0057] In the above S204, at least one first agent component corresponding to the first subtask can be understood as the agent component required to execute the first subtask.

[0058] In one implementation, S204 may include the following steps: for each first subtask, the task information of the first subtask is matched with the functional information of each intelligent agent component in the intelligent agent component library to obtain the matching degree between each intelligent agent component and the first subtask; at least one intelligent agent component whose matching degree with the first subtask satisfies the preset matching condition is obtained from the intelligent agent component library and is used as at least one first intelligent agent component corresponding to the first subtask.

[0059] More specifically, as mentioned above, the first process may include nodes of the types such as agent nodes, logic nodes, input nodes, and output nodes. Since the first subtasks corresponding to agent nodes and logic nodes can be implemented by agent components, for such first subtasks, their task information can be matched with the functional information of each agent component to determine the first agent component that corresponds to such first subtasks.

[0060] Matching the task information of the first subtask with the functional information of the agent components can be achieved in various ways. For example, based on the task information such as the name, description, and expected input / output of the first subtask, a third natural language description can be generated; the semantic similarity between the third natural language description and the functional information of each agent component can be calculated as the matching degree between the two.

[0061] The preset matching conditions can be set according to actual needs, such as the matching degree being greater than or equal to a preset threshold, etc., but this application embodiment does not limit this.

[0062] In the above implementation, standardized component management improves the reusability of intelligent agent components and reduces the resource overhead required for creating intelligent assistants.

[0063] S206, assemble at least one first intelligent agent component corresponding to the first subtask in the first process to obtain the first intelligent agent corresponding to the first subtask in the first process, and create the interaction logic between the first intelligent agents corresponding to the first subtask based on the logical relationship between the first subtasks to obtain the first intelligent assistant.

[0064] The logical relationships between the first subtasks can include, but are not limited to, at least one of the following: the execution order of the first subtasks (e.g., serial or parallel), dependencies (e.g., a subtask depends on the execution result of another subtask), etc. Based on this, after obtaining at least one first agent component corresponding to the first subtask, the calling order between the first agents corresponding to the first subtask can be created based on the execution order; alternatively, dependencies can be created between the first agents corresponding to the first subtask based on the dependencies, such as the input of one first agent depending on the output of another first agent, etc. Thus, the creation of the interaction logic between the first agent components corresponding to the first subtask is completed.

[0065] For example, the interaction logic between first-level agents includes the following two types: Pipeline-style interaction logic: First agents are invoked sequentially. The output of the previous first agent serves as the input of the next. For example, the output of the information retrieval agent serves as the input of the content generation agent, and the output of the content generation agent serves as the input of the quality verification agent. This interaction logic is suitable for linear, branchless first-stage processes.

[0066] The tree-branching interaction logic: Based on the condition judgments of the logical nodes in the first process, the first process is split into multiple parallel or serial sub-branches, and at least one first agent is called according to the sub-branch. This interaction logic is suitable for tasks that require dynamic adjustment.

[0067] The first intelligent assistant may include the first intelligent agent corresponding to the first subtask in the first process and the interaction logic between these first intelligent agents. The first intelligent assistant can receive relevant parameters input by the user, process the relevant parameters based on the first process, and obtain the processing result. Specifically, the first intelligent assistant can, based on relevant parameters and interaction logic, call the first intelligent agent to execute the first subtask in the first process.

[0068] The intelligent assistant creation method provided in this embodiment integrates a large number of process templates through a process template library. During the creation process, a matching first process template is obtained from the template library according to the task requirements, and a first process executed by the created intelligent assistant is generated based on this template. This reduces the randomness in the process generation process and ensures that the generated first process fully meets the task requirements. Furthermore, a large number of intelligent agent components are integrated through an intelligent agent component library. Based on the description information of each intelligent agent component, at least one first intelligent agent component is matched for the first subtask in the first process. By assembling these first intelligent agent components, the first intelligent agent corresponding to the first subtask is obtained. Based on the logical relationship between the first subtasks, the interaction logic between the first intelligent agents corresponding to the first subtasks is created, and a first intelligent agent assistant that fully meets the requirements of the first task can be quickly obtained. It is evident that users only need to input the task requirements for creating a smart assistant to trigger a fully automated smart assistant creation process, significantly lowering the barrier to entry for smart assistant creation and reducing the requirement for users' professional knowledge, enabling non-professional users to easily complete the creation of a smart assistant; secondly, the fully automated process efficiently and conveniently creates smart assistants that meet user needs, and the smart assistant can be rebuilt simply by modifying the task requirements, thus better adapting to the diversification and dynamic evolution of user needs.

[0069] In some other embodiments, after S206 described above, the following may also be included: S208, based on the first process, tests the first intelligent assistant and obtains the test results.

[0070] In one implementation, the first intelligent assistant can undergo both local (node-level) and overall (process-level) usability verification based on the first process. The test result is determined based on the verification results of these two types of usability verification. Local usability verification refers to verifying whether the first intelligent agent within the first intelligent assistant correctly executes its corresponding first sub-task. Overall usability verification refers to simulating a real-world task scenario, inputting complete task requirements, and verifying whether the interaction logic between the various first intelligent agents is smooth and whether the first intelligent assistant correctly executes the first process.

[0071] Specifically, S208 may include: for each first subtask in the first process, in response to the first subtask having a corresponding first agent, instructing the first agent to execute the first subtask and obtaining the execution result of the first subtask; instructing the first intelligent assistant to execute the first process and obtaining the execution result of the first process; and determining the test result of the first intelligent assistant based on the execution result of the first subtask and the execution result of the first process.

[0072] More specifically, as mentioned above, the first process may include nodes of various types such as agent nodes, logic nodes, input nodes, and output nodes. Since the first subtasks corresponding to agent nodes and logic nodes can be implemented by the first agent, for such first subtasks, the corresponding first agent can be instructed to execute such first subtasks to obtain the execution results of such first subtasks.

[0073] For the first subtask corresponding to the agent node, verify whether the execution result of the first subtask conforms to the functional information of each first agent component in the corresponding first agent; if yes, it indicates that the first intelligent assistant has passed the local availability verification of the agent node; if no, it indicates that the first intelligent assistant has failed the local availability verification of the agent node.

[0074] For the first subtask corresponding to the logical node, it determines whether the conditional logic triggers the correct branch process; if yes, it indicates that the first intelligent assistant has passed the local availability verification of the logical node; if no, it indicates that the first intelligent assistant has failed the local availability verification of the logical node.

[0075] Furthermore, the execution result of the first process can be compared with the expected execution result of the first process. If the two are consistent, it indicates that the first intelligent assistant has passed the overall usability verification; if the two are inconsistent, it indicates that the first intelligent assistant has not passed the overall usability verification.

[0076] Furthermore, in response to the first intelligent assistant passing both partial and overall usability verification, the test result is determined to be "test passed"; in response to the first intelligent assistant failing either partial or overall usability verification, the test result is determined to be "test failed".

[0077] In practice, during the partial availability verification process, the first intelligent agent in the first intelligent assistant can be instructed to execute the corresponding first sub-task by providing the first intelligent assistant with the corresponding prompt word; during the overall availability verification process, the first intelligent assistant can also be instructed to execute the first process by providing the first intelligent assistant with the corresponding prompt word.

[0078] This implementation method can accurately test whether the first intelligent assistant executes the first process correctly, avoid logical errors or intelligent component call failures during the use of the first intelligent assistant, enable users to discover and fix problems in time before use, and improve the reliability of the first intelligent assistant.

[0079] In another implementation, in S208 above, only partial availability verification or overall availability verification can be performed on the first intelligent assistant.

[0080] The foregoing illustrates a partial implementation of S208. It should be understood that S208 can also be implemented in other ways, and this application embodiment does not limit this.

[0081] Based on test results, the S210 optimizes the first intelligent assistant.

[0082] In one implementation, in response to a test result indicating that the test failed, the above steps S202 to S208 are re-executed until the test result of the first intelligent assistant indicates that the test passed.

[0083] In another implementation, in response to a test result indicating that the test has passed, a first intelligent assistant is presented; in response to receiving negative feedback from the user on the first intelligent assistant, a first prompt message is presented, which is used to indicate optimization suggestions for the first intelligent assistant; in response to receiving the first optimization suggestion message input by the user, the first task requirement is updated based on the first optimization suggestion, and based on the updated first task requirement, the steps of obtaining a first process template matching the first task requirement from the template library and presenting the created first intelligent assistant are repeated until a positive feedback message from the user on the created first intelligent assistant is received, that is, the above steps S202~S210 are repeated.

[0084] Users can provide feedback to the first intelligent assistant by rating it. For example, if the user's score is greater than a preset score, it means the user is relatively satisfied with the first intelligent assistant, and a positive feedback operation is confirmed. If the user's score is less than or equal to the preset score, it means the user is dissatisfied with the first intelligent assistant, and a negative feedback operation is confirmed.

[0085] In other embodiments, the method may further include: in response to receiving a positive feedback operation from a user to a first intelligent assistant, writing a first process executed by the constructed first intelligent assistant as a process template into a template library. This enables dynamic updating of the template library.

[0086] As can be seen, this implementation method uses a closed-loop mechanism of user feedback to use test results for real-time optimization of task requirements and reasoning planning. Through user feedback optimization, the created first intelligent assistant continuously meets the real needs of users, thus achieving dynamic optimization of the first intelligent assistant.

[0087] The intelligent assistant creation method provided in this embodiment realizes an automated closed-loop process of "task requirement input → automatic generation of the first process → creation of intelligent agent component device and intelligent assistant → verification and testing → intelligent assistant optimization", which can further improve the matching degree between the created intelligent assistant and the user's real needs.

[0088] In some other embodiments, after S208 described above, the method may further include: in response to a test result indicating that the test has passed, writing the first process as a process template into a template library. This enables dynamic updating of the template library.

[0089] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0090] Based on the same inventive concept, this application also provides an intelligent assistant creation system. Please refer to... Figure 3 This is a schematic diagram of the structure of an intelligent assistant creation system 300 provided in an embodiment of this application. The system includes: a demand processing intelligent agent subsystem, a reasoning and planning intelligent agent subsystem, an assembly and coordination subsystem, and a data pool. The data pool includes a template library and an intelligent agent component library.

[0091] The demand processing intelligent agent subsystem is used to obtain the first task demand for creating an intelligent assistant from the user's instructions. The reasoning and planning intelligent agent subsystem is used to respond to the first task requirement, obtain a first process template matching the first task requirement from the template library, and generate a first process executed by the created intelligent assistant based on the first process template; the first process includes multiple first sub-tasks and the logical relationship between the multiple first sub-tasks. The assembly and coordination subsystem is used to match at least one first intelligent agent component for the first subtask in the first process based on the description information of each intelligent agent component in the intelligent agent component library, and to assemble at least one first intelligent agent component corresponding to the first subtask in the first process to obtain the first intelligent agent corresponding to the first subtask in the first process. Based on the logical relationship, it creates the interaction logic between the first intelligent agents corresponding to the first subtask in the first process to obtain the first intelligent assistant.

[0092] In other embodiments, the demand processing intelligent agent subsystem is used for: Receive the dialogue data input by the user in the dialogue interface; The dialogue data is subjected to intent recognition to obtain a first intent; In response to the first intent to create a smart assistant, the first task requirement is generated based on the first intent.

[0093] In other embodiments, the data pool also includes a user feature library; The demand processing intelligent agent subsystem generates the first task demand in the following manner: In response to the fact that the dialogue data does not contain the user's first user feature, the user's second user feature is obtained from the user feature library, and the first intent is expanded based on the second user feature to obtain the first task requirement; In response to the fact that the dialogue data contains the user's first user feature, the first intent is expanded based on the first user feature to obtain the first task requirement, and the first user feature is written into the user feature library to update the second user feature.

[0094] In other embodiments, the reasoning and planning agent subsystem is used for: The first task requirement is broken down into multiple second sub-tasks; The multiple second subtasks and the subtasks in each process template in the template library are matched to obtain the matching degree between the first task requirements and each process template. Based on the matching degree between the first task requirement and each process template, a first process template that matches the first task requirement in the template library is determined.

[0095] In other embodiments, the description information of the intelligent agent component includes the functional information of the intelligent agent component; The assembly and coordination subsystem is used for: For each first subtask, the task information of the first subtask is matched with the functional information of each intelligent agent component in the intelligent agent component library to obtain the matching degree between each intelligent agent component and the first subtask. At least one intelligent agent component that meets the preset matching conditions with the first subtask is obtained from the intelligent agent component library and is used as at least one first intelligent agent component corresponding to the first subtask.

[0096] In other embodiments, the intelligent assistant creation system further includes a test intelligent agent subsystem; The test intelligent agent subsystem is used for: The first intelligent assistant was tested based on the first process, and the test results were obtained; Based on the test results, the first intelligent assistant is optimized.

[0097] In other embodiments, the test intelligent agent subsystem optimizes the first intelligent assistant in the following ways: In response to the test result being passed, the first intelligent assistant is displayed; In response to receiving negative feedback from the user on the first smart assistant, a first prompt message is presented, which is used to indicate optimization suggestions for the first smart assistant; In response to receiving the first optimization suggestion information input by the user, the first task requirement is updated based on the first optimization suggestion, and based on the updated first task requirement, the step of obtaining the first process template matching the first task requirement from the template library and presenting the created first smart assistant is repeatedly triggered until the positive feedback operation of the user on the created first smart assistant is received.

[0098] In other embodiments, the test intelligent agent subsystem tests the first intelligent assistant in the following manner: For each first subtask in the first process, in response to the first subtask having a corresponding first agent, the first agent is instructed to execute the first subtask and obtain the execution result of the first subtask; Instruct the first intelligent assistant to execute the first process and obtain the execution result of the first process; Based on the execution results of the first subtask and the execution results of the first process, the test results of the first intelligent assistant are determined.

[0099] In this embodiment, the reasoning planning agent subsystem can be a series system of a question-answering agent and a prompt word generation agent, both of which are embedded with RAG technology.

[0100] The question-answering agent is used to respond to the first task requirement, obtain a first process template that matches the first task requirement from the template library, and generate a first process to be executed by the created intelligent assistant based on the first process template.

[0101] The prompt word generating agent is used to generate corresponding prompt word templates for the first agent corresponding to the first subtask. The prompt word templates are used to generate prompt words provided to the first agent so that the first agent can better understand and execute the corresponding first subtask. For example, such as... Figure 4 As shown, the prompt word generation agent receives all agent node information and logic node information specified in the main body of the first process, and generates dedicated prompt word templates for them to ensure that the underlying model components can accurately understand the subtask objective. Conditional judgment rules can be defined in the prompt word templates corresponding to the logic nodes.

[0102] The template library, agent component library, and user feature library in the data pool have already been described in the method implementation section above and will not be repeated here. The structure of the data pool is as follows: Figure 5 As shown.

[0103] It's worth noting that all data within the data pool supports visualization, specifically in three aspects: The visualization of the intelligent agent component library enables a structured presentation of all components, facilitating user operations such as adding, deleting, querying, and modifying; the visualization of the template library automatically maps JSON-formatted process templates to intelligent agent collaboration flowcharts, allowing users to easily adjust workflow logic via drag-and-drop, with the adjustments synchronously updated in the corresponding process templates; and the visualization of the task completion process and results refers to the retention and display of information when the created intelligent assistant completes user tasks, including interaction information between various intelligent agent components within the intelligent assistant and the final task completion result.

[0104] To facilitate understanding of the interaction process between the various subsystems and the data pool in the intelligent assistant creation system, the following will combine... Figure 6 The process will be explained.

[0105] like Figure 6 As shown, the demand processing intelligent agent subsystem performs intent recognition on the dialogue data input by the user, and dynamically fills and expands the identified first intent by combining it with the user's user features in the user feature library to obtain the first task demand, and sends the first task demand to the reasoning and planning intelligent agent subsystem for processing.

[0106] The reasoning and planning intelligent agent subsystem retrieves a matching first process template from the template library based on the first task requirements, generates a first process to be executed by the created intelligent assistant based on the first process template, and sends the first process to the assembly and collaboration subsystem for processing.

[0107] The assembly and coordination subsystem matches the first intelligent agent component corresponding to the first subtask in the first process from the intelligent agent component library, and obtains the first intelligent agent corresponding to the first subtask by assembling the first intelligent agent component; further, based on the logical relationship between the first subtasks, the interaction logic between the first intelligent agents is created, thus obtaining the first intelligent assistant.

[0108] The testing intelligent agent subsystem performs partial and overall usability verification on the first intelligent assistant. In response to the first intelligent assistant passing the test, it presents the assistant to the user and receives user feedback. Upon receiving positive user feedback, it writes the first process as a user history template into the template library. Upon receiving negative user feedback, it presents a first prompt message, instructing the user to input optimization suggestions for the first intelligent assistant. Based on the user's first optimization suggestion, the requirements processing intelligent agent subsystem updates the first task requirements and then re-executes the steps of creating the first intelligent assistant, testing the created assistant, and optimizing it through human-computer interaction collaboration, until the created first intelligent assistant passes the test and receives positive user feedback.

[0109] The intelligent assistant creation system provided in this application, through a modular multi-agent subsystem collaboration mechanism, eliminates the reliance on a fully manual first-process configuration. Users only need to input task requirements to trigger the system to automatically generate the first process. Simultaneously, the first process is iteratively optimized based on user feedback, significantly simplifying the intelligent assistant creation process and allowing even non-professional users to easily create intelligent assistants. Furthermore, this system includes a data pool containing a user feature library, a template library, and an intelligent agent component library. Multiple subsystems interact with the data pool to automatically obtain suitable user features, process templates, and intelligent agent components. This reduces the randomness of the first-process generation process, lowers the creation difficulty and resource consumption, and ensures that the created intelligent assistant fully meets the user's actual needs by accurately capturing those needs.

[0110] This application also proposes a computer-readable storage medium that stores one or more programs, the programs including instructions that, when executed by an electronic device including multiple applications, enable the electronic device to perform... Figure 2 The method of the illustrated embodiment is specifically used to perform the following operations: In response to a user's instruction to create a smart assistant, a first process template matching the first task requirement is retrieved from a template library, and a first process executed by the created smart assistant is generated based on the first process template; the first process includes multiple first sub-tasks and the logical relationships between the multiple first sub-tasks. Based on the description information of each intelligent agent component in the intelligent agent component library, at least one first intelligent agent component is matched for the first subtask in the first process; Assemble at least one first intelligent agent component corresponding to the first subtask in the first process to obtain the first intelligent agent corresponding to the first subtask in the first process, and based on the logical relationship, create the interaction logic between the first intelligent agents corresponding to the first subtask in the first process to obtain the first intelligent assistant.

[0111] In summary, the above description is merely a preferred embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

[0112] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0113] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (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 optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0114] 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 a process, method, article, or apparatus. Without further limitation, 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 said element.

[0115] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

Claims

1. A method for creating an intelligent assistant, characterized in that, include: In response to the user's instruction to create a smart assistant, a first process template matching the first task requirement is obtained from the template library, and a first process is generated based on the first process template to be executed by the created smart assistant. The first process includes multiple first subtasks and the logical relationships between the multiple first subtasks; Based on the description information of each intelligent agent component in the intelligent agent component library, at least one first intelligent agent component is matched for the first subtask in the first process; Assemble at least one first intelligent agent component corresponding to the first subtask in the first process to obtain the first intelligent agent corresponding to the first subtask in the first process, and based on the logical relationship, create the interaction logic between the first intelligent agents corresponding to the first subtask in the first process to obtain the first intelligent assistant.

2. The method according to claim 1, characterized in that, The first task requirement was obtained in the following way: Receive the dialogue data input by the user in the dialogue interface; The dialogue data is subjected to intent recognition to obtain a first intent; In response to the first intent to create a smart assistant, the first task requirement is generated based on the first intent.

3. The method according to claim 2, characterized in that, The step of generating the first task requirement based on the first intent includes: In response to the fact that the dialogue data does not contain the user's first user feature, the user's second user feature is obtained from the user feature library, and the first intent is expanded based on the second user feature to obtain the first task requirement; In response to the fact that the dialogue data contains the user's first user feature, the first intent is expanded based on the first user feature to obtain the first task requirement, and the first user feature is written into the user feature library to update the second user feature.

4. The method according to claim 1, characterized in that, The template library is used to obtain a first process template that matches the requirements of the first task, including: The first task requirement is broken down into multiple second sub-tasks; The multiple second subtasks and the subtasks in each process template in the template library are matched to obtain the matching degree between the first task requirements and each process template. Based on the matching degree between the first task requirement and each process template, a first process template that matches the first task requirement in the template library is determined.

5. The method according to claim 1, characterized in that, The description information of the intelligent agent component includes the functional information of the intelligent agent component; The method of matching at least one first intelligent agent component to the first subtask in the first process based on the description information of each intelligent agent component in the intelligent agent component library includes: For each first subtask, the task information of the first subtask is matched with the functional information of each intelligent agent component in the intelligent agent component library to obtain the matching degree between each intelligent agent component and the first subtask. At least one intelligent agent component that meets the preset matching conditions with the first subtask is obtained from the intelligent agent component library and is used as at least one first intelligent agent component corresponding to the first subtask.

6. The method according to claim 1, characterized in that, The method further includes: The first intelligent assistant was tested based on the first process, and the test results were obtained; Based on the test results, the first intelligent assistant is optimized.

7. The method according to claim 6, characterized in that, The optimization of the first intelligent assistant based on the test results includes: In response to the test result being passed, the first intelligent assistant is displayed; In response to receiving negative feedback from the user on the first smart assistant, a first prompt message is presented, which is used to indicate optimization suggestions for the first smart assistant; In response to receiving the first optimization suggestion information input by the user, the first task requirement is updated based on the first optimization suggestion, and based on the updated first task requirement, the steps of obtaining the first process template matching the first task requirement from the template library and presenting the created first smart assistant are repeated until positive feedback from the user on the created first smart assistant is received.

8. The method according to claim 6, characterized in that, The step of testing the first intelligent assistant based on the first process to obtain test results includes: For each first subtask in the first process, in response to the first subtask having a corresponding first agent, the first agent is instructed to execute the first subtask and obtain the execution result of the first subtask; Instruct the first intelligent assistant to execute the first process and obtain the execution result of the first process; Based on the execution results of the first subtask and the execution results of the first process, the test results of the first intelligent assistant are determined.

9. A smart assistant creation system, characterized in that, include: The system includes a demand processing intelligent agent subsystem, a reasoning and planning intelligent agent subsystem, an assembly and coordination subsystem, and a data pool, wherein the data pool contains a template library and an intelligent agent component library. The demand processing intelligent agent subsystem is used to obtain the first task demand for creating an intelligent assistant from the user's instructions. The reasoning and planning intelligent agent subsystem is used to respond to the first task requirement, obtain a first process template that matches the first task requirement from the template library, and generate a first process to be executed by the created intelligent assistant based on the first process template. The first process includes multiple first subtasks and the logical relationships between the multiple first subtasks; The assembly and coordination subsystem is used to match at least one first intelligent agent component for the first subtask in the first process based on the description information of each intelligent agent component in the intelligent agent component library, and to assemble at least one first intelligent agent component corresponding to the first subtask in the first process to obtain the first intelligent agent corresponding to the first subtask in the first process. Based on the logical relationship, it creates the interaction logic between the first intelligent agents corresponding to the first subtask in the first process to obtain the first intelligent assistant.

10. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the smart assistant creation method as described in any one of claims 1 to 8.

Images