Task execution method, system, apparatus, device, and storage medium

Abstract

This application provides a task execution method, system, apparatus, device, and storage medium. The method includes: in response to receiving a task execution request, invoking a task control module to determine the application product identifier corresponding to the task execution request; if the task execution mode of the task to be executed is local execution, invoking a task execution module matched to the task to be executed to obtain target AI configuration information mapped to the application product identifier from a preset AI configuration information database; executing the task to be executed based on the target AI configuration information; and outputting the task execution result. This implementation eliminates the need for redundant development, offers good scalability, facilitates the reuse of intelligent service functions of the task execution module, and saves AI resources.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence (AI) technology, and in particular to a task execution method, system, apparatus, device, and storage medium. Background Technology

[0002] With the diversification of tasks in applications, for application scenarios involving multiple application products, each application product is typically based on an independently deployed intelligent robot. Any intelligent robot can respond to the task execution information of its corresponding application product and return the task execution result for that application product. Specifically, the intelligent robot can access its AI configuration information (such as prompts, vector knowledge bases, etc.), execute tasks based on the task execution logic corresponding to that application product, and obtain the task execution result.

[0003] In conventional implementations, the AI ​​configuration information and task execution logic required for any intelligent robot to perform a task are encapsulated within the intelligent robot instance. That is, the intelligent robots corresponding to multiple application products are independent of each other. As a result, every time a new application product is added, the entire intelligent robot development and configuration process needs to be repeated for that application product. This not only wastes AI resources and has poor scalability, but also results in high maintenance costs because the AI ​​configuration information is embedded in various intelligent robot instances. Summary of the Invention

[0004] To overcome the problems existing in related technologies, embodiments of this application provide a task execution method, system, apparatus, device, and storage medium.

[0005] According to a first aspect of the embodiments of this application, a task execution method is provided, the method comprising: Upon receiving a task execution request, the task control module is invoked to determine the application product identifier corresponding to the task execution request; When the task execution mode of the task to be executed is local execution, the task execution module matching the task to be executed is invoked to obtain the target AI configuration information mapped to the application product identifier from the preset artificial intelligence (AI) configuration information database. The task to be executed is performed based on the target AI configuration information, and the task execution result is output.

[0006] According to a second aspect of the embodiments of this application, a task execution system is provided, the system including a task control module and multiple task execution modules, wherein the system maintains a preset artificial intelligence (AI) configuration information database. The task decomposition intelligent module is used to respond to receiving a task execution request by calling the task control module to determine the application product identifier corresponding to the task execution request; and to determine the task execution mode of the task to be executed. The task execution module matching the task to be executed is used to obtain target AI configuration information mapped to the application product identifier from the AI ​​configuration information database when the task execution mode of the task to be executed is local execution; and to execute the task to be executed based on the target AI configuration information, and output the task execution result.

[0007] According to a third aspect of the embodiments of this application, a task execution apparatus is provided, the apparatus comprising: The determining unit is used to, in response to receiving a task execution request, call the task control module to determine the application product identifier corresponding to the task execution request; The acquisition unit is used to, when the task execution mode of the task to be executed is local execution, call the task execution module matched with the task to be executed to obtain the target AI configuration information mapped to the application product identifier from the preset artificial intelligence AI configuration information database; The task execution unit is used to execute the task to be executed based on the target AI configuration information and to output the task execution result.

[0008] According to a fourth aspect of the embodiments of this application, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method as described in the first aspect.

[0009] According to a fifth aspect of the embodiments of this application, a computer-readable storage medium is provided having a computer program stored thereon, the program being executed by a processor to implement the method as described in the first aspect.

[0010] According to a sixth aspect of the embodiments of this application, a computer program product is provided, including a computer program that is executed by a processor to implement the method of the first aspect.

[0011] The technical solutions provided in this application embodiment may include the following beneficial effects: The local task execution system has a pre-set AI configuration information database. Each AI configuration information in the database is mapped to an application product identifier. Therefore, the technical solution of this application embodiment maintains the AI ​​configuration information of each application product uniformly within the task execution system. Compared to the decentralized nature of AI configuration information, this reduces the maintenance cost of the AI ​​configuration information for each application product. Upon receiving any task execution request, the task control module can determine the application product identifier corresponding to the task execution request. Then, it calls the task execution module matching the task to be executed to obtain the target AI configuration information mapped to the application product identifier, and executes the task based on the target AI configuration information. In other words, this technical solution, by marking and driving AI configuration information with application product identifiers, abandons the model of "one set of AI configuration information and task execution module per application product." When adding a new application product, the task execution logic of the new application product is added by setting the AI ​​configuration information of the new application product mapped to the application product identifier in the task execution system, eliminating the need for repeated development and providing good scalability. Furthermore, for any task to be executed, dynamically selecting the matching AI configuration information through the application product identifier facilitates the reuse of the intelligent service functions of the task execution module and saves AI resources.

[0012] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly described below. It should be understood that those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0014] Figure 1 This application provides an exemplary task execution system architecture diagram. Figure 2 An exemplary method flow diagram of the task execution method provided in the embodiments of this application; Figure 3 This application provides an exemplary data flow diagram among multiple intelligent modules. Figure 4 This is an exemplary schematic diagram of the task execution device provided in the embodiments of this application; Figure 5 This is an exemplary structural diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

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

[0016] The terminology used in the following embodiments of this application is for the purpose of describing specific embodiments and is not intended to limit the technical solutions of this application. As used in the specification and appended claims of this application, the singular expressions "a," "an," "the," "the," "the," and "this" are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that although the terms first, second, etc., may be used in the following embodiments to describe a class of objects, the objects are not limited to these terms. These terms are used to distinguish specific implementations of that class of objects.

[0017] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0018] This application's embodiments relate to the deployment of online consultation robots for various application products and the execution technology of consultation tasks in task scenarios involving multiple application products. Taking a financial application scenario as an example, with the diversification of financial-related tasks, financial applications can deploy multiple application products, such as "instant repayment," "merchant loans," and "freight insurance." To provide intelligent online consultation services for each application product, an intelligent robot is typically deployed for each application product. Each intelligent robot can respond to the consultation tasks of its corresponding application product and return the execution results of the corresponding consultation tasks. Due to the significant differences in task execution logic, knowledge systems, and service interfaces among different application products, to achieve precise service, a dedicated intelligent customer service robot is typically deployed independently for each application product. Each intelligent robot instance is independently configured with its corresponding application product's matching knowledge base, intent recognition model, and AI configuration information and task execution logic, such as interaction logic with backend tasks, to adapt to the service needs of that application product.

[0019] For example, the process of creating a separate intelligent robot instance for the application product "Instant Payment" may include: independently writing the intent recognition logic and dialogue flow for "Instant Payment" matching; independently configuring the knowledge base and question-and-answer information pairs for "Instant Payment"; independently integrating the interface call logic between the intelligent robot and the backend "Instant Payment" task system; and independently configuring the logging, monitoring, and auditing schemes for the intelligent robot. Creating a separate intelligent robot instance for another application product (such as merchant loans) requires repeating the above development process. It is evident that this approach of deploying intelligent robots independently for each application product leads to isolation between the intelligent robots. This not only prevents the sharing of core intelligent module capabilities and knowledge bases, wasting AI resources, but also results in poor scalability. Furthermore, the dispersed embedding of AI configuration information across various intelligent robot instances leads to high maintenance costs.

[0020] In view of this, the technical solution provided in this application maintains an AI configuration information database for all application products in the task execution system and maps application product identifiers (productTags) to corresponding AI configuration information. This allows for unified maintenance of AI configuration information for each application product within the task execution system. Upon receiving any task execution request, the system searches for the mapped target AI configuration information based on the application product identifier corresponding to the task execution request. Then, multiple intelligent modules within the task execution system are invoked to execute the task based on the target AI configuration information. Therefore, this technical solution, by uniformly maintaining AI configuration information and managing various intelligent modules, and using application product identifiers to mark and drive AI configuration information, allows for the addition of task execution logic for new application products by setting the AI ​​configuration information of the new application product mapped to the application product identifier in the task execution system. This eliminates the need for repetitive development, provides good scalability, and facilitates the reuse of intelligent service functions of the task execution module and saves AI resources by dynamically selecting the matching AI configuration information for the task to be executed using the application product identifier.

[0021] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0022] The task execution method provided in this application embodiment can be applied to a task execution system, which can be deployed in electronic devices such as computers, game consoles, or personal computers (PCs), or in cloud computers, distributed clusters, etc. According to its functions and calling logic, the task execution system in this application embodiment can be represented as a layered architecture, with each layer deploying corresponding functional modules and performing corresponding functional divisions. For example... Figure 1 As shown, the task execution system may include an access layer, an execution layer, a configuration layer, and a management layer, wherein the layers can communicate with each other through an application programming interface (API).

[0023] The access layer can serve as the network entry point and traffic gateway for the task execution system, used to receive, parse, and initially distribute task execution requests from external clients (such as merchant front-end applications). The access layer can provide a user-facing input interface to receive task instructions expressed by the user in natural language or audio. In some embodiments, an information receiving client and an information analysis module can be deployed in the access layer. The information receiving client can be implemented as a visual input interface, which, after receiving the text of the task execution request, can transmit the text to the information analysis module. The information analysis module parses the text of the task execution request to obtain the application product identifier corresponding to the task execution request. Then, by analyzing the application product identifier, it determines the execution mode suitable for the task execution request and routes the text of the task execution request to the intelligent module in the execution layer that matches the execution mode.

[0024] For example, the information receiving client can transmit text to the information analysis module via the HTTP protocol and push task execution requests using Server-Sent Events (SSE) streaming technology, thereby improving the response speed of task execution requests. The information analysis module can then extract and parse application product identifiers or task keywords from the request text to determine the appropriate execution mode for the task execution request.

[0025] The execution layer can serve as the control center of a task execution system, coordinating and scheduling other intelligent modules within the system, executing task processing flows, and making critical decisions. For example, the execution layer can be used for intent recognition and routing of task execution request text, managing local and third-party execution modes, and scheduling multiple task execution modules. In some embodiments, the execution layer can deploy a scheduling service module, an intent recognition module, a routing module, a question-and-answer task processing module, a tool invocation module, and a third-party AI access module. The scheduling service module and the third-party AI access module are task execution modules under two different execution modes. In one implementation scenario, either the scheduling service module or the third-party AI access module receives a request from the information analysis module and then invokes subsequent intelligent modules according to the corresponding task execution logic. The third-party AI access module is a task execution module under the third-party execution mode, also known as an external execution path task execution module. After receiving a request from the information analysis module, the third-party AI access module can directly invoke a third-party AI platform to execute tasks. The scheduling service module is the task execution module in local execution mode, also known as the task execution module of the internal execution path. After receiving a request from the information analysis module, the scheduling service module can call the intent recognition module to determine the type of the task to be executed, and call the routing module to dispatch the task execution request to the intelligent module with the matching type, so as to call the corresponding intelligent module to execute the task. In some embodiments, the scheduling service module is also used to call the services of the management layer to ensure that the dialogue information during the task execution process is recorded.

[0026] The intent recognition module can be used to obtain the intent recognition prompt corresponding to the application product identifier corresponding to the task execution request from the configuration layer, and then identify the task type corresponding to the request text based on the intent recognition prompt. Then, it can output the task type identifier to the routing module. The task type identifier can be, for example, Frequently Asked Questions (FAQ) and TOOL_REQUIRED.

[0027] The routing module can be used to determine the flow of tasks based on the recognition results of the intent recognition module. For example, the routing module can route tasks to be executed with the task type marked as FAQ to the question and answer task processing module, and route tasks to be executed with the task type marked as TOOL_REQUIRED to the tool invocation module.

[0028] The question-and-answer task processing module (FAQ module) can be used to process static task-type question-and-answer tasks, which refer to question-and-answer tasks that rely on a static knowledge base such as the product's definition, rules, processes, status, and operation methods. In some embodiments, the question-and-answer task processing module can obtain the question-and-answer prompts corresponding to the application product identifier from the configuration layer, and then retrieve the answer information matching the task execution request from a preset question-and-answer knowledge base based on the question-and-answer prompts.

[0029] The tool invocation module can be used to handle dynamic question-and-answer tasks. Dynamic question-and-answer tasks refer to tasks with personalized intent characteristics. These personalized intent characteristics can include tasks with dynamically changing states and user-specific unique tasks, such as the task "Why hasn't this order been paid?" or the task "Check my account's coupon limit." In some embodiments, the tool invocation module can obtain tool configuration information corresponding to the application product identifier from the configuration layer. Then, it can invoke the tool to execute the task based on the tool configuration information.

[0030] The configuration layer can be the AI ​​asset configuration layer of the task execution system. It is used to configure various drivers of the task execution system and maintain and manage various AI configuration information. For example, the configuration layer can configure the operating logic and data supporting the operation of various functional modules in the access layer and execution layer. It can also manage and maintain intent recognition prompt word libraries, question-and-answer prompt word libraries, knowledge bases, tool configuration information libraries, etc. The configuration layer can also be used to store the configuration information of a new product after the task execution system receives it, and to establish a mapping relationship between the application product identifier of the new product and the configuration information of the new product.

[0031] In some embodiments, the configuration layer may include a prompt word management module, a knowledge base, and a tool configuration information management module. The prompt word management module manages various prompt words corresponding to all application product identifiers, such as intent recognition prompt words, question-and-answer prompt word libraries, and tool invocation instructions. The knowledge base manages and caches vector knowledge bases corresponding to each application product identifier, so that the question-and-answer task processing module can retrieve response information matching the task execution request. The tool configuration information management module manages the tool configuration information of tools that the task execution system can call. The tool configuration information may include tool call interface addresses and call parameters, and the tool configuration information management module is also used to provide the required target tool configuration information to the tool calling module.

[0032] The management layer can be a functional layer for monitoring and auditing session information in the task execution system. It can be used for system-wide session management, and all execution records of tasks to be executed for any of the above execution paths are forcibly stored to ensure that all dialogue interactions are traceable and measurable. In some embodiments, a session management module and an audit module can be deployed in the management layer. The session management module is used to persistently store the complete dialogue record of each task execution request, including user questions, AI responses, application product identifiers, timestamps, processing paths, and executed modules, providing a data foundation for problem tracing, effect analysis, and compliance auditing. The audit module is used to statistically analyze parameters such as the number of calls, success rate, and time consumption of each application product identifier, each functional module, and each path, providing data support for task execution decisions.

[0033] In this way, regardless of whether the task to be executed is executed through the internal or external path of the task execution system, it can be monitored and statistically analyzed by this task execution system. In particular, it can effectively avoid scenarios where calling third-party AI platforms to execute tasks is easy to bypass monitoring, thereby meeting the requirements of traceable dialogue records and auditable operations.

[0034] It should be noted that all modules in the functional layers of the task execution system can be intelligent modules. Any intelligent module can be implemented as a functional module with natural language understanding and task execution capabilities, such as an AI agent. An intelligent module is a dynamic integration of model functionality and task execution capabilities. It can understand received natural language instructions, determine the action to be executed based on the understanding, and then execute the corresponding action. In some embodiments, the prompt word can be text or a sentence input to the large language model, used to guide the product recommendation model to generate response information or as the starting point for continuing text generation. The prompt word can be at least one of a sentence, a question, a descriptive paragraph, and a series of instructions. The intelligent module executes the task according to the guidance of the corresponding prompt word. As an example, the intent recognition prompt word can guide the intent recognition module to parse the semantics of the task to be executed to identify the task type, thereby assigning the task to a matching task execution module and triggering the task execution module to execute the task.

[0035] As can be seen, this embodiment maintains an AI configuration information database for all application products and various intelligent modules within the task execution system. It maps application product identifiers to corresponding AI configuration information. Upon receiving any task execution request, it invokes multiple intelligent modules within the task execution system, driven by the application product identifier, to execute the task based on the target AI configuration information corresponding to the application product identifier. On one hand, adding task execution logic to new application products eliminates the need for repetitive development, offering good scalability. The method of dynamically selecting matching AI configuration information for the task to be executed via application product identifier facilitates the reuse of intelligent service functions of the task execution module, saving AI resources.

[0036] It should be understood that the above description of each functional layer of the task execution system is merely illustrative and does not constitute a limitation on the functional layers of the task execution system in this application embodiment. In actual implementation scenarios, the task execution system may include more or fewer functional layers, and each layer may contain more or fewer functional modules. This application embodiment does not limit this.

[0037] Please refer to Figure 2 , Figure 2 An exemplary task execution method is shown, which can be performed by... Figure 1 The illustrated task execution system is being executed. This task execution method may include the following steps: In step S101, in response to receiving a task execution request, the task control module is invoked to determine the application product identifier corresponding to the task execution request.

[0038] In step S102, if the task execution mode of the task to be executed is local execution, the task execution module matching the task to be executed is invoked to obtain the target AI configuration information mapped to the application product identifier from the preset AI configuration information library.

[0039] In step S103, the task to be executed is performed based on the target AI configuration information, and the task execution result is output.

[0040] The application product identifier can be a string formed by a number sequence, code, hash value, etc. Each application product identifier can uniquely identify an application product to distinguish different application products. For example, the application product identifier for the application product "disburse" is "disburse", and the application product identifier for the application product "loan" is "loan". For application products that require a third-party AI platform, the application product identifier is, for example, "custom_ai".

[0041] The AI ​​configuration information database includes multiple AI configuration information entries, which are the AI ​​configuration information corresponding to the configured application products. In some embodiments, AI configuration information refers to the configurable resources pre-configured in the task execution system to support the functions of intelligent modules. During task execution, AI configuration information can be loaded and injected into the corresponding intelligent modules to drive them to complete specific tasks such as intent recognition, knowledge retrieval, question-and-answer generation, or tool invocation. For example, AI configuration information may include intent recognition prompts, question-and-answer prompt word libraries, tool invocation instructions, vector knowledge bases, and tool configuration information. In this embodiment, the AI ​​configuration information database is uniformly maintained by the task execution system. The AI ​​configuration information of different application products is bound to the application product identifier of that application product, enabling each intelligent module in the task execution system to dynamically invoke the matching AI configuration information based on the application product identifier.

[0042] As can be seen, the technical solution of this application embodiment, compared with the decentralized AI configuration information, helps to reduce the maintenance cost of AI configuration information for each application product. Moreover, by marking and driving AI configuration information with application product identifiers, the model of "one set of AI configuration information and task execution module for each application product" is abandoned. When adding a new application product, the task execution logic of the new application product is added by setting the AI ​​configuration information of the new application product mapped by the application product identifier in the task execution system, and the capabilities of the intelligent module are transformed from "hard-coded" to "configurable". This achieves "zero-code" configuration of new products, with good scalability. In addition, for any task to be executed, the method of dynamically selecting the AI ​​configuration information matching the task to be executed through the application product identifier is conducive to reusing the intelligent service functions of the task execution module and saving AI resources.

[0043] In some implementation scenarios, the information receiving client may include options for different types of questions. The question types can be categorized according to application products; for example, each question option may correspond to an application product identifier. In some embodiments, in response to a user selecting any question type during the task execution request input phase, the task execution request may carry the application product identifier corresponding to the selected question type. In some embodiments, in response to a user not selecting a question type during the task execution request input phase, the task execution request does not carry an application product identifier.

[0044] Based on this, in response to receiving a task execution request, the task control module can parse whether the task execution request contains an application product identifier. If the task execution request contains an application product identifier, it means that the user selected a question type during the task execution request input stage, and the application product identifier can be read directly. If the task execution request does not contain an application product identifier, it means that the user did not select a question type during the task execution request input stage. The task control module can identify task keywords in the text of the task execution request and determine the application product identifier corresponding to the task keywords based on a preset identifier mapping relationship, which is then used as the application product identifier. The identifier mapping relationship includes multiple application product identifiers and a keyword library corresponding to each application product identifier.

[0045] For example, the keyword library corresponding to the application product identifier "Disburse" includes keywords such as "repayment," "credit limit," "early payment," "loan disbursement," and "arrival time." The keyword library corresponding to the application product identifier "Loan" includes keywords such as "loan," "borrow money," "interest," "interest rate," "credit limit," and "repayment." If the task execution request text is, for example, "I want to know how the early payment credit limit is calculated," and parsing the request determines that it does not contain an application product identifier, the task control module identifies the task keywords in the text "I want to know how the early payment credit limit is calculated." For example, the identified task keywords include "early payment" and "credit limit." Since both "early payment" and "credit limit" match the keyword library of the application product identifier "disburse," it can be determined that the application product identifier corresponding to the text "I want to know how the early payment credit limit is calculated" is Disburse.

[0046] Furthermore, the task execution system supports two execution modes for the task to be executed: local execution (i.e., internal path execution) and execution via a third-party AI platform (i.e., external path execution). In some embodiments, after determining the application product identifier corresponding to the task execution request, the information analysis module in the task execution system can determine the task execution mode matching the application product identifier based on preset matching rules. If the task execution mode is third-party execution, the third-party AI access module is invoked to access the third-party AI platform, so that the third-party AI platform executes the task to be executed. If the task execution mode of the task to be executed is local execution, step S102 above is executed.

[0047] The preset matching rules maintain the matching relationship between application product identifiers and execution modes. For example, the execution mode matched by the application product identifier `custom_ai` is the execution mode of a third-party AI platform, while the execution modes matched by the application product identifiers `disburse` and `loan` are local execution modes. After determining the execution mode corresponding to the task to be executed, the information analysis module can route the text of the task execution request to the corresponding functional module. For instance, if the application product identifier is `custom_ai`, and the execution mode corresponding to the task to be executed is determined to be third-party AI platform execution, the text of the task execution request will be routed to... Figure 1 The third-party AI access module illustrated in the diagram, when the application product identifier is Disburse, determines that the execution mode corresponding to the task to be executed is local execution, and routes the text of the task execution request to... Figure 1 The diagram illustrates the scheduling service module.

[0048] As can be seen, the technical solution of this application embodiment adopts a dual-mode mechanism to support the coexistence of local execution mode and third-party AI access mode in a task execution system, which is beneficial to the compatibility of local execution mode and third-party AI access mode.

[0049] In some embodiments, after receiving the text of the task execution request, the third-party AI access module can directly call the API interface of the third-party AI platform through the pre-configured adapter in the task execution system, thereby calling the large model of the third-party AI platform to execute the task to be executed in this application embodiment. Given the diversity of application products and the complexity of user problems, the task types to be executed can be divided into static task types and dynamic task types according to task attributes and the processing logic of tasks with different attributes. Static task types and dynamic task types correspond to different task processing paths and are executed by different intelligent modules. For example, the identifier for a static task type is, for example, FAQ, and the identifier for a dynamic task type is, for example, TOOL_REQUIRED. For a detailed description of static and dynamic task types, please refer to the description of the above embodiments, which will not be repeated here.

[0050] Based on this, after receiving the text of the task execution request, the scheduling service module calls the task type recognition module to obtain the target type recognition prompt word marked by the application product identifier from the pre-configured type recognition prompt word library. Based on the target type recognition prompt word, the target task type of the task to be executed is identified. The target task type is either a static task type or a dynamic task type. Then, the task execution module matching the target task type is called to obtain the target AI configuration information from the AI ​​configuration information library corresponding to the target task type.

[0051] Combination Figure 1In the illustrative embodiment, the task type identification module can be an intent identification module. For example, the scheduling service module can call the intent identification module to obtain the target type identification prompt words marked by the application product identifier. Then, the intent identification module can identify the intent representation of the text of the task execution request based on the target type identification prompt words. If the intent representation does not contain personalized intent features, the target task type of the task to be executed is determined to be the static task type. If the intent representation contains the personalized intent features, the target task type of the task to be executed is determined to be the dynamic task type. Here, personalized intent features refer to tasks with dynamically changing states and / or unique tasks dependent on the user. Intent representations that do not contain personalized intent features refer to intent representations dependent on the product's definition, rules, processes, states, operation methods, etc.

[0052] Among them, target type identification prompts are, for example, intent identification prompts. For instance, the guidance logic for intent identification prompts is as follows: “ Please determine which of the following categories the user's question belongs to: -FAQ: Questions can be answered as follows: "What is Instant Repayment?", "Why has my withdrawable amount for Instant Repayment decreased?", "Can I make a repayment myself?", "How do I activate it?", "What are repayment limits and repayment amounts?", "How do I check these orders?", "How do I activate Instant Repayment?", "Why doesn't the manual repayment service fee match the payment?", "Where can I view the Instant Repayment transaction history?", "How do I view the details of my credit limit usage?" Questions raised included: "Where is the instant payment entry?", "Order confirmation did not release credit limit", "What is the instant payment repayment ratio?", "How long do I need to operate before I can activate instant payment?", "Can I repay offline if I can't repay online?", "How do I change the receiving account for instant payment?", "Early repayment for instant payment", "Is the credit limit not restored after confirming receipt of instant payment?", "Why was my deposit deducted?", "Why didn't my instant payment limit change after shipment?", "Why is my manual payment only 17, but switching to automatic payment gives me 9200? Why is there such a big difference?", and "The instant payment entry is gone." -TOOL_REQUIRED: The issue requires access to system tools to obtain real-time information, such as "What is my credit limit?", "Did I receive a payment yesterday?", "Why wasn't this order paid out?", "I haven't activated it?", "Why was my account frozen?", "Why was my instant payment service frozen?", "When transferring store ownership, the message 'Instant payment service is not activated,' what does this mean?", "Issue type: Instant payment"; -OTHER: Other unrecognized issues.

[0053] Please return only one tag, without adding any explanation.

[0054] Example 1: User: "What is the express payment service?" Output: FAQ Example 2: User: "What is my credit limit?" Output: TOOL_REQUIRED " In this way, by representing the user intent through the question text of the task execution request, it is easier to accurately dispatch the task to the matching processing path, thereby facilitating efficient and accurate automated task processing.

[0055] In some embodiments, if the target task type of the task to be executed is a static task type, it is only necessary to search for the answer information of the question corresponding to the task execution request in the knowledge base. Based on this, it can be determined that the task execution module is a question-and-answer task processing module, and the routing module is called to route the task execution request to the question-and-answer task processing module. Then, the question-and-answer task processing module is called to obtain the target question-and-answer prompt word corresponding to the application product identifier from the pre-configured question-and-answer prompt word library as the target AI configuration information, and to obtain the question-and-answer information matching the target question-and-answer prompt word from the pre-configured knowledge base. The task execution result is generated based on the answer information in the question-and-answer information.

[0056] The target question-answering prompt can be a pre-configured Retrieval-Augmented Generation (RAG) prompt. RAG prompts are a technique that combines vector retrieval with natural language responses generated by a large model. RAG prompts can guide the question-answering task processing module to first retrieve question-answer information pairs related to the task execution request from the vector knowledge base corresponding to the application product identifier, and then integrate the retrieved question-answer information to generate a response that is more in line with the task execution request.

[0057] The pre-configured knowledge base can be implemented as a vector knowledge base (VectorStore). A vector knowledge base is a knowledge base obtained by converting task description documents (such as product manuals and operation guides) input during the application product configuration phase into feature vectors. Each application product's vector knowledge base is bound to its application product identifier. In some embodiments, the pre-configured knowledge base may also include general question-and-answer knowledge applicable to multiple application products; this general question-and-answer knowledge may not be bound to any application product identifier.

[0058] For example, obtaining question-and-answer information matching the target question-and-answer prompt from a pre-configured knowledge base can be achieved by performing semantic similarity retrieval on vector knowledge in the pre-configured knowledge base according to the guidance of the target question-and-answer prompt, so as to obtain the vector knowledge with the highest similarity to the question corresponding to the task execution request.

[0059] For example, the text corresponding to the task execution request is "What is the repayment ratio for 'Instant Repayment'?", and the corresponding application product identifier is Disburse. After determining that the task execution request is a static task type, the question-and-answer task processing module can be called to process this task. The target question-and-answer prompt bound to Disburse is, for example, "You are an intelligent question-and-answer assistant for the 'Instant Repayment' task. Please answer the user's question in a professional, concise, and friendly manner, strictly based on the knowledge provided below. If the answer is not explicitly contained in the knowledge base, please inform the user 'No relevant information was found; it is recommended to transfer to a human agent.'" Then, by searching in the vector knowledge base, the question-and-answer knowledge "The highest repayment ratio currently supported by 'Instant Repayment' is 90% of the order amount; the actual ratio will be dynamically adjusted based on the merchant's historical credit" is matched. This answer information can then be fed back to the client for display to the user.

[0060] It should be noted that the question-and-answer information matching the target question-and-answer prompt may include at least one. In some embodiments, when there is only one question-and-answer information matching the target question-and-answer prompt, the question-and-answer task processing module can extract the response information from that question-and-answer information and provide feedback to the client. In some embodiments, when there are at least two question-and-answer information matching the target question-and-answer prompt, the question-and-answer task processing module can extract the response information from the at least two question-and-answer information, and then, according to the semantic understanding of the task execution request, integrate the extracted at least two response information into a response information feedback client for the task execution request.

[0061] In some embodiments, if the target task type of the task to be executed is the dynamic task type, it can be determined that the task to be executed is related to at least one dynamic factor among user account, time, and task status. Since a task matching response cannot be obtained through a general knowledge base, a task execution tool should be invoked for real-time processing to obtain an accurate task processing result. Based on this, it can be determined that the task execution module is a tool invocation module. The invocation module routes the task execution request to the tool invocation module, which then retrieves the tool configuration information corresponding to the application product identifier from a pre-configured tool information database as the target AI configuration information. Subsequently, the tool configuration information can be sent to the task server corresponding to the application product identifier, so that the task server can invoke the tool to execute the task to be executed based on the tool configuration information. It should be understood that the task server is a third-party service system, and the task execution system can also invoke the task server via an adapter.

[0062] The tool configuration information corresponding to the application product identifier may include the tool identifier, the tool call interface address, and the call input parameters.

[0063] For example, the text corresponding to the task execution request is "What is my instant repayment limit?", and the application product identifier corresponding to the task execution request is Disburse. After determining that the task execution request is a dynamic task type, the tool invocation module can be called to query the tool configuration information that has a mapping relationship with Disburse from the tool configuration information management module. The tool configuration information includes, for example, the tool name queryDisburseLimit, the API endpoint https: / / servers.xxxxx.com / api / disburse / limit, the calling method GET, parameter mapping rules, etc. Then, the tool invocation module generates an HTTP request based on the task execution request and the tool configuration information, and calls the task server to execute the corresponding task via the API endpoint. After obtaining the response information from the task server, key data is extracted from the response information according to the parameter mapping rules in the tool configuration information, and the task execution result is generated according to the semantics of the task execution request. For example, the key data in the response information from the task server is a limit of 5000. For "What is my instant repayment limit?", the task execution result "Hello, your current available instant repayment limit is 5,000 yuan" can be generated.

[0064] As can be seen, this implementation method, for tasks executed locally, divides task types and execution resources into static and dynamic tasks, thereby decoupling standardized services from personalized services and reducing the complexity of the task execution system. Furthermore, accurately determining the task processing method according to user intent and calling the matching processing path not only reduces the complexity of task processing logic but also improves the accuracy and timeliness of response information.

[0065] In some embodiments, for any of the modules among the question-and-answer task processing module, tool invocation module, and third-party AI access module, the task execution system calls the session management module to store the session context corresponding to the task to be executed and the application product identifier. When the audit conditions are met, multiple session contexts corresponding to each application product identifier are audited.

[0066] For example, the session management module can use persistent storage to store the complete dialogue information for each task. The complete dialogue information may include the task execution request text, the task execution result, the application product identifier, the timestamp, etc.

[0067] In some embodiments, the auditing module determines whether audit conditions are met, and if so, audits multiple session contexts corresponding to each application product identifier. For example, meeting audit conditions could be reaching a preset audit duration, or reaching a preset number of session contexts corresponding to an application product identifier. The audit duration could be, for example, 5 hours, and the preset number of session contexts could be, for example, 1000. For example, the auditing module can audit parameters such as call volume, intent distribution, response time, and task execution success rate of multiple session contexts corresponding to the same application product identifier.

[0068] This not only supports unified management and flexible expansion of related tasks for multiple application products, but also ensures the traceability of session contexts for various processing paths by intercepting the session context of all task processing paths.

[0069] In some embodiments, at any execution stage of the above embodiments, the task execution system can receive a configuration instruction from a configuration client for a new application product. The configuration instruction includes an application product identifier for the new application product and AI configuration information for the new application product. The AI ​​configuration information includes at least one of the following: a type identification prompt word corresponding to the new application product, a question-and-answer prompt word corresponding to the new application product, a knowledge base corresponding to the new application product, and tool configuration information corresponding to the new application product. Furthermore, the task execution system can establish a mapping relationship between the application product identifier of the new application product and the AI ​​configuration information of the new application product, so that during the execution of tasks related to the new application product, the intelligent module in the task execution system can call the AI ​​configuration information of the new application product to perform the task.

[0070] As can be seen, due to the technical solution of this application, the AI ​​configuration information of each application product is uniformly maintained in the task execution system. The application product identifier is used to bind the corresponding AI configuration information by mapping the application product identifier. When a new application product is added, the task execution logic of the new application product is added by setting the AI ​​configuration information of the new application product through the mapping of the application product identifier in the task execution system. There is no need for repeated development, and the scalability is good.

[0071] The above is a description of the embodiments of this application from the perspective of a task execution system. The following will discuss... Figure 1 The technical solution of the embodiments of this application is described from the perspective of mutual cooperation between intelligent modules illustrated in the figure.

[0072] refer to Figure 3 , Figure 3 This paper illustrates an exemplary data flow among multiple intelligent modules. The following section describes the data flow involved in the task execution process of this application embodiment from the perspective of each task execution path.

[0073] Step S31: The information receiving client receives the product inquiry request input by the user and sends the product inquiry request to the information analysis module.

[0074] The product consultation request is implemented as a natural language command. This product consultation request can trigger the task execution system to execute a consultation task. Therefore, the product consultation request is the aforementioned task execution request.

[0075] In step S32, the information analysis module determines the execution mode of the product consultation request based on the productTag of the product consultation request. If it is determined to be an internal execution mode, the product consultation request is routed to the scheduling service module and step S33 is executed; if it is an external execution mode, the product consultation request is routed to the third-party AI access module and step S34 is executed.

[0076] In step S33, the scheduling service module obtains the target intent prompt word corresponding to the productTag from the prompt word management module, and calls the intent recognition module to identify the task type of the product consultation request based on the target intent prompt word. If it is determined to be a static task type, step S331 is executed; if it is determined to be a dynamic task type, step S333 is executed.

[0077] Step S331: The routing module is invoked to route the product consultation request to the FAQ module, entering the first internal execution path.

[0078] In step S332, the FAQ module determines the question and answer prompts corresponding to the productTag from the question and answer prompt word library, retrieves the answer information from the knowledge base based on the question and answer prompts, and generates the product consultation result, i.e., the task execution result.

[0079] Step S333: The routing module is invoked to route the product inquiry request to the tool invocation module, thus entering the second internal execution path.

[0080] In step S334, the tool invocation module determines the target tool configuration information corresponding to the productTag from the tool configuration information database, and calls the external task server to execute the task based on the target tool configuration information in order to receive the product consultation results returned by the external task server.

[0081] In step S34, the third-party AI access module calls the third-party AI to perform the task and receives the product consultation results returned by the third-party AI.

[0082] During the execution phase of the above steps, the session management module obtains the session context information of the product consultation request from the FAQ module, tool call module or third-party AI access module. Then, the audit module audits the multiple session contexts corresponding to the productTag.

[0083] In summary, in the technical solution of this application embodiment, an AI configuration information database is preset in the local task execution system. Each AI configuration information in the database is mapped to an application product identifier. Therefore, the technical solution of this application embodiment maintains the AI ​​configuration information of each application product uniformly within the task execution system. Compared to the decentralized nature of AI configuration information, this reduces the maintenance cost of the AI ​​configuration information for each application product. Upon receiving any task execution request, the task control module can determine the application product identifier corresponding to the task execution request. Then, it calls the task execution module matching the task to be executed to obtain the target AI configuration information mapped to the application product identifier, and executes the task based on the target AI configuration information. In other words, this technical solution, by using product identification tags and driving AI configuration information, abandons the model of "one set of AI configuration information and task execution modules for each application product". When adding a new application product, the task execution logic of the new application product is added by setting the AI ​​configuration information of the new application product mapped by the application product identification in the task execution system, without the need for repeated development, and has good scalability. In addition, for any task to be executed, the method of dynamically selecting the AI ​​configuration information matching the task to be executed by using the application product identification is conducive to reusing the intelligent service functions of the task execution module and saving AI resources.

[0084] The above embodiments describe various implementations of the method provided in this application from the perspectives of determining the application product identifier, determining the execution mode based on the application product identifier, identifying the type of task to be executed, and establishing a mapping relationship between AI configuration information and the application product identifier. It should be understood that, corresponding to the above processing steps, the embodiments of this application can implement the above functions in hardware or a combination of hardware and computer software. Whether a function is executed by hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0085] For example, if the above implementation steps are achieved through software modules to realize the corresponding functions. Figure 4 As shown, a task execution device is provided, which may include a determining unit 41, an acquiring unit 42, and a task execution unit 43. This task execution device can be deployed in... Figure 1 In the illustrated task execution system, the above is used to perform... Figures 2 to 3 Some or all of the operations in the process.

[0086] For example: the determining unit 41 is used to, in response to receiving a task execution request, call the task control module to determine the application product identifier corresponding to the task execution request; the obtaining unit 42 is used to, when the task execution mode of the task to be executed is local execution, call the task execution module matched to the task to be executed to obtain the target AI configuration information mapped to the application product identifier from the preset artificial intelligence AI configuration information database; the task execution unit 43 is used to execute the task to be executed based on the target AI configuration information and output the task execution result.

[0087] Optionally, the determining unit 41 is further configured to parse whether the task execution request contains an application product identifier; if the task execution request contains an application product identifier, read the application product identifier; the determining unit 41 is further configured to, if the task execution request does not contain an application product identifier, identify task keywords in the text of the task execution request; determine the application product identifier corresponding to the task keyword based on a preset identifier correspondence relationship, and use it as the application product identifier, wherein the identifier correspondence relationship includes multiple application product identifiers and a keyword library corresponding to each application product identifier.

[0088] Optionally, the task execution device further includes an access unit and a determination unit 41, which is further configured to determine the task execution mode matching the application product identifier based on a preset matching rule; the access unit is configured to call a third-party AI access module to access a third-party AI platform when the task execution mode is third-party execution, so that the third-party AI platform can execute the task to be executed.

[0089] Optionally, the acquisition unit 42 is further configured to call the task type recognition module to obtain the target type recognition prompt word marked by the application product identifier from the pre-configured type recognition prompt word library, identify the target task type of the task to be executed based on the target type recognition prompt word, wherein the target task type is a static task type or a dynamic task type; and call the task execution module matching the target task type to obtain the target AI configuration information from the AI ​​configuration information library corresponding to the target task type.

[0090] Optionally, the acquisition unit 42 is further configured to identify the intent representation of the text of the task execution request; if the intent representation does not contain personalized intent features, determine that the target task type of the task to be executed is the static task type; the acquisition unit 42 is further configured to determine that the target task type of the task to be executed is the dynamic task type if the intent representation contains the personalized intent features.

[0091] Optionally, the acquisition unit 42 is further configured to determine that the task execution module is a question-and-answer task processing module if the target task type of the task to be executed is the static task type; and call the question-and-answer task processing module to obtain the target question-and-answer prompt words corresponding to the application product identifier from the pre-configured question-and-answer prompt word library as the target AI configuration information.

[0092] Optionally, the task execution unit 43 is further configured to obtain question and answer information matching the target question and answer prompt from a pre-configured knowledge base; and generate the task execution result based on the response information in the question and answer information.

[0093] Optionally, the acquisition unit 42 is further configured to determine that the task execution module is a tool invocation module if the target task type of the task to be executed is the dynamic task type; and to invoke the tool invocation module to obtain the tool configuration information corresponding to the application product identifier from the pre-configured tool information library as the target AI configuration information.

[0094] Optionally, the task execution unit 43 is further configured to send the tool configuration information to the task server corresponding to the application product identifier, so that the task server can call the tool to execute the task to be executed based on the tool configuration information.

[0095] Optionally, the task execution device further includes a storage unit and an auditing unit. The storage unit is used to call the session management module to store the session context corresponding to the task to be executed and the application product identifier. The auditing unit is used to audit multiple session contexts corresponding to each application product identifier when the auditing conditions are met.

[0096] Optionally, the task execution device further includes a receiving unit and a mapping establishment unit. The receiving unit is used to receive a configuration instruction for a new application product. The configuration instruction includes an application product identifier of the new application product and AI configuration information of the new application product. The AI ​​configuration information includes at least one of the following: a type identification prompt word corresponding to the new application product, a question-and-answer prompt word corresponding to the new application product, a knowledge base corresponding to the new application product, and tool configuration information corresponding to the new application product. The mapping establishment unit is used to establish a mapping relationship between the application product identifier of the new application product and the AI ​​configuration information of the new application product.

[0097] It is understandable that the above division of units is only a logical functional division. In actual implementation, the functions of the above units can be integrated into hardware entities. For example, the functions of the determination unit 41, the acquisition unit 42, and the task execution unit 43 can be integrated into the processor, and the functions of the receiving unit can be integrated into the transceiver.

[0098] Please refer to Figure 5 , Figure 5 An electronic device 6 is provided, which can be implemented as carrying Figure 1 The device is a task execution system within a computer. This electronic device may include a processor 61, a transceiver 62, and a memory 63, these components being connected and communicating via a communication bus 64. The transceiver 62 can be used for communication to transmit various task-related natural language instructions and information. The memory 63 is used to store computer programs and information generated during task execution; when the computer program is invoked, it causes the processor 61 to execute the aforementioned commands. Figures 2 to 3 The operation of some or all of the electronic devices.

[0099] For details on the implementation process, please refer to [link / reference]. Figures 2 to 3 The relevant descriptions are omitted here.

[0100] It should be understood that, in some feasible implementations, Figure 5 The processor illustrated can be a central processing unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. Figure 5 The memory illustrated may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of the memory may also include non-volatile random access memory. For example, the memory may also store information such as the device type.

[0101] This application also provides a computer-readable storage medium storing image processing instructions, which, when executed on a computer, cause the computer to perform the aforementioned actions. Figures 2 to 3 Some or all of the steps in the method described in the illustrated embodiments.

[0102] This application also provides a computer program product including instructions for task execution, which, when run on a computer, causes the computer to perform the aforementioned... Figures 2 to 3 Some or all of the steps in the method described in the illustrated embodiments.

[0103] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0105] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0106] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0107] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, smartphone, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0108] Although alternative embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make further changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0109] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of this application should be included within the scope of protection of this invention.

Claims

1. A task execution method characterized by, The method includes: Upon receiving a task execution request, the task control module is invoked to determine the application product identifier corresponding to the task execution request; When the task execution mode of the task to be executed is local execution, the task execution module matching the task to be executed is invoked to obtain the target AI configuration information mapped to the application product identifier from the preset artificial intelligence (AI) configuration information database. The task to be executed is performed based on the target AI configuration information, and the task execution result is output.

2. The method of claim 1, wherein, The task control module determines the application product identifier corresponding to the task execution request, including: Analyze whether the task execution request contains an application product identifier; If the task execution request contains an application product identifier, read the application product identifier; If the task execution request does not contain an application product identifier, identify the task keywords in the text of the task execution request; The application product identifier corresponding to the task keyword is determined based on a preset identifier correspondence relationship. The identifier correspondence relationship includes multiple application product identifiers and a keyword library corresponding to each application product identifier.

3. The method of claim 1, wherein, After the task control module determines the application product identifier corresponding to the task execution request, the process also includes: The task execution mode for matching the application product identifier is determined based on preset matching rules; When the task execution mode is third-party execution, the third-party AI access module is invoked to access the third-party AI platform so that the third-party AI platform can execute the task to be executed.

4. The method according to any one of claims 1 to 3, characterized in that, The step of invoking the task execution module matching the task to be executed, and obtaining the target AI configuration information mapped to the application product identifier from the preset AI configuration information database, includes: The task type recognition module is invoked to retrieve the target type recognition prompt word labeled with the application product identifier from the pre-configured type recognition prompt word library. The target task type of the task to be executed is identified based on the target type identification prompt, and the target task type is either a static task type or a dynamic task type; The task execution module matching the target task type is invoked to obtain the target AI configuration information from the AI ​​configuration information library corresponding to the target task type.

5. The method of claim 4, wherein, The process of identifying the target task type of the task to be executed based on the target type identification prompt includes: Identify the intent representation of the text of the task execution request; If the intent representation does not contain personalized intent features, the target task type of the task to be executed is determined to be the static task type. If the intent representation includes the personalized intent feature, the target task type of the task to be executed is determined to be the dynamic task type.

6. The method according to claim 4 or 5, characterized in that, The step of calling the task execution module that matches the target task type to obtain the target AI configuration information from the AI ​​configuration information database corresponding to the target task type includes: If the target task type of the task to be executed is the static task type, then the task execution module is determined to be the question-and-answer task processing module. The question-and-answer task processing module is invoked to retrieve the target question-and-answer prompt words corresponding to the application product identifier from the pre-configured question-and-answer prompt word library, which are used as the target AI configuration information.

7. The method of claim 6, wherein, The execution of the task to be executed based on the target AI configuration information includes: Retrieve question and answer information that matches the target question and answer prompt from a pre-configured knowledge base; The task execution result is generated based on the response information in the question-and-answer information.

8. The method according to claim 4 or 5, characterized in that, The step of calling the task execution module that matches the target task type to obtain the target AI configuration information from the AI ​​configuration information database corresponding to the target task type includes: If the target task type of the task to be executed is the dynamic task type, then the task execution module is determined to be a tool invocation module; The tool invocation module is invoked to retrieve the tool configuration information corresponding to the application product identifier from the pre-configured tool information library, which is then used as the target AI configuration information.

9. The method according to claim 8, characterized in that, The execution of the task to be executed based on the target AI configuration information includes: The tool configuration information is sent to the task server corresponding to the application product identifier, so that the task server can call the tool to execute the task to be executed based on the tool configuration information.

10. The method according to any one of claims 1-3, characterized in that, The method further includes: The session management module is invoked to store the corresponding session context and application product identifier of the task to be executed. If the audit criteria are met, audit multiple session contexts corresponding to each application product identifier.

11. The method according to any one of claims 1-3, characterized in that, The method further includes: Receive configuration instructions for a new application product. The configuration instructions include the application product identifier of the new application product and the AI ​​configuration information of the new application product. The AI ​​configuration information includes at least one of the following: type recognition prompt words corresponding to the new application product, question and answer prompt words corresponding to the new application product, knowledge base corresponding to the new application product, and tool configuration information corresponding to the new application product. Establish a mapping relationship between the application product identifier of the new application product and the AI ​​configuration information of the new application product.

12. A task execution system, characterized in that, The system includes a task control module and multiple task execution modules. The system maintains a preset AI configuration information database. The task decomposition intelligent module is used to respond to receiving a task execution request by calling the task control module to determine the application product identifier corresponding to the task execution request; and to determine the task execution mode of the task to be executed. The task execution module matching the task to be executed is used to obtain target AI configuration information mapped to the application product identifier from the AI ​​configuration information database when the task execution mode of the task to be executed is local execution; and to execute the task to be executed based on the target AI configuration information, and output the task execution result.

13. A task execution device, characterized in that, The device includes: The determining unit is used to, in response to receiving a task execution request, call the task control module to determine the application product identifier corresponding to the task execution request; The acquisition unit is used to, when the task execution mode of the task to be executed is local execution, call the task execution module matched with the task to be executed to obtain the target AI configuration information mapped to the application product identifier from the preset artificial intelligence AI configuration information database; The task execution unit is used to execute the task to be executed based on the target AI configuration information and to output the task execution result.

14. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the program to implement the method as described in any one of claims 1-11.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by a processor to implement the method as described in any one of claims 1-11.

16. A computer program product, comprising a computer program, characterized in that, The computer program is executed by a processor to implement the method according to any one of claims 1-11.

Images