Task execution method and device, storage medium and electronic equipment

Abstract

The embodiment of the invention provides a task execution method and device, a storage medium and electronic equipment, and the method comprises the steps: recognizing the attribute information of a target task, splitting the target task based on the attribute information, and obtaining N target sub-tasks; determining M sub-network models according to the sub-task types of the N target sub-tasks; executing a test result obtained by the test set through the M sub-network models, and screening out P target sub-network models from the M sub-network models according to the test result; and executing the N target sub-tasks according to the target execution sequence through the P target sub-network models. The problem that the task cannot be effectively executed in the prior art is solved, and the effect of effectively executing the task is achieved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of computer technology, and more specifically, to a method and device for executing a task, a storage medium, and an electronic device. Background Art

[0002] Traditional model-as-a-service functions usually predefine the task execution process manually. In this model, the execution effect of the task is highly dependent on the designer's expertise and experience. However, due to the limitations of individual cognition and the limited scope of information reception, this method of manually defining the process is difficult to fully tap the potential of generative artificial intelligence in content generation. In addition, once the task execution process and model selection are fixed, it is difficult to perform automatic tuning or model replacement even in the face of changing environments or task requirements, and there is a problem of inability to effectively execute tasks. Summary of the invention

[0003] The embodiments of the present application provide a method and device for executing a task, a storage medium, and an electronic device to at least solve the problem that tasks cannot be effectively executed in the related art.

[0004] According to an embodiment of the present application, a task execution method is provided, including: identifying attribute information of a target task to split the target task based on the attribute information to obtain N target subtasks, wherein N is a natural number greater than or equal to 1; determining M subnetwork models according to the subtask types of the N target subtasks, wherein M is a natural number greater than or equal to 1, the subnetwork model is used to execute the target subtask, and one subtask type corresponds to at least one subnetwork model; obtaining test results by executing a test set through the M subnetwork models, and screening out P target subnetwork models from the M subnetwork models according to the test results, wherein P is a natural number less than or equal to N, one subtask type corresponds to one test set, and one test set corresponds to at least one subnetwork model; executing the N target subtasks through the P target subnetwork models and in a target execution order, wherein the target execution order is an execution order determined based on the attribute information and historical execution results.

[0005] In an exemplary embodiment, before identifying the attribute information of the target task and splitting the target task based on the attribute information to obtain N target subtasks, the method further includes: acquiring task information and context information of the target task, wherein the task information is used to represent the function implemented by the target task, and the context information is used to represent the situational background of the target task; and determining the attribute information of the target task from the task information and the context information.

[0006] In an exemplary embodiment, the attribute information of the target task is identified to split the target task based on the attribute information to obtain N target subtasks, including: splitting the functions implemented by the target task based on the classification of the attribute information, and combining the functions implemented by the split target tasks according to the functions implemented by the target task to obtain multiple function combinations; screening out a target function combination from the multiple function combinations based on the historical execution results, wherein the target function combination is a function combination whose historical execution results meet the preset execution results among the multiple function combinations; splitting the target task according to the target function combination to obtain N target subtasks.

[0007] In an exemplary embodiment, before executing the N target subtasks through the P target sub-network models and in the target execution order, the method further includes: respectively determining the K target sub-functions implemented by the N target subtasks and the N groups of sub-attribute information corresponding to the N target subtasks, wherein the K is a natural number greater than or equal to the N; determining the target dependency relationship between the N target subtasks according to the K target subfunctions, wherein the target dependency relationship is used to represent the individual steps of executing the N target tasks and the association relationship between the individual steps; determining the initial execution order of executing the N target subtasks from the target dependency relationship; adjusting the initial execution order based on the historical execution results and the N groups of sub-attribute information to obtain the target execution order.

[0008] In an exemplary embodiment, the test results obtained by executing the test set through the M above-mentioned sub-network models, and P target sub-network models are screened out from the M above-mentioned sub-network models according to the above-mentioned test results, including: inputting the above-mentioned test set into the M above-mentioned sub-network models, so as to perform the test operation on the above-mentioned test set through the M above-mentioned sub-network models, and obtain the M above-mentioned test results; calculating the performance indicators of the M above-mentioned sub-network models through the matching results between the above-mentioned test results and the preset test results, wherein the above-mentioned preset test results are pre-set results based on the functions implemented by the above-mentioned test set; according to the above-mentioned performance indicators and the above-mentioned historical execution results, screening out P above-mentioned target sub-network models from the M above-mentioned sub-network models, wherein the above-mentioned target sub-network model is a model in which the above-mentioned performance indicators meet the preset screening conditions among multiple sub-network models, and the above-mentioned historical execution results meet the preset execution results.

[0009] In an exemplary embodiment, determining the target dependency between the N target subtasks mentioned above according to the K target subfunctions mentioned above includes: determining the target information for realizing the K target subfunctions mentioned above, wherein the target information includes input and output information, resource information, time information and priority information when executing the target subtasks, the input and output information is used to determine the input and output dependencies for executing the N target subtasks mentioned above, the resource information is used to determine the resource dependencies for executing the N target subtasks mentioned above, and the time information and the priority information are used to determine the execution dependencies for executing the N target subtasks mentioned above; and determining the input and output dependencies, the resource dependencies and the execution dependencies as the target dependencies.

[0010] In an exemplary embodiment, after executing N of the above-mentioned target subtasks through P of the above-mentioned target sub-network models and in a target execution order, the above-mentioned method further includes: collecting feedback information when the above-mentioned target sub-network model executes the above-mentioned target subtask during the execution process, wherein the above-mentioned feedback information is used to indicate the execution status of the above-mentioned target subtask; collecting result information when the above-mentioned target sub-network model executes the above-mentioned target subtask after the execution is completed; and determining the above-mentioned feedback information and the above-mentioned result information as the target execution result.

[0011] According to an embodiment of the present application, a task execution device includes: a first identification module, which is used to identify attribute information of a target task, so as to split the target task based on the attribute information to obtain N target subtasks, wherein the N is a natural number greater than or equal to 1; a first determination module, which is used to determine M subnetwork models according to the subtask types of the N target subtasks, wherein the M is a natural number greater than or equal to 1, the subnetwork model is used to execute the target subtask, and one subtask type corresponds to at least one subnetwork model; a first screening module, which is used to obtain test results by executing a test set through the M subnetwork models, and to screen out P target subnetwork models from the M subnetwork models according to the test results, wherein the P is a natural number less than or equal to the N, one subtask type corresponds to one test set, and one test set corresponds to at least one subnetwork model; a first execution module, which is used to execute the N target subtasks through the P target subnetwork models and in a target execution order, wherein the target execution order is an execution order determined based on the attribute information and historical execution results.

[0012] In an exemplary embodiment, the above-mentioned device also includes: a first acquisition module, used to identify the attribute information of the target task, so as to split the above-mentioned target task based on the above-mentioned attribute information, and before obtaining N target subtasks, acquire the task information and context information of the above-mentioned target task, wherein the above-mentioned task information is used to represent the function implemented by the above-mentioned target task, and the above-mentioned context information is used to represent the situational background of the above-mentioned target task; a second determination module, used to determine the attribute information of the above-mentioned target task from the above-mentioned task information and the above-mentioned context information.

[0013] In an exemplary embodiment, the above-mentioned first identification module includes: a first splitting sub-module, which is used to split the functions implemented by the above-mentioned target tasks based on the classification of the above-mentioned attribute information, and split the functions implemented by the above-mentioned target tasks according to the function combinations implemented by the above-mentioned target tasks, to obtain multiple function combinations; a first screening sub-module, which is used to screen out a target function combination from the multiple function combinations based on the above-mentioned historical execution results, wherein the above-mentioned target function combination is a function combination whose historical execution results meet the preset execution results among the multiple function combinations; a second splitting sub-module, which is used to split the above-mentioned target task according to the above-mentioned target function combination, to obtain N of the above-mentioned target sub-tasks.

[0014] In an exemplary embodiment, the above-mentioned device also includes: a third determination module, which is used to determine the K target sub-functions implemented by the N above-mentioned target sub-tasks and the N groups of sub-attribute information corresponding to the N above-mentioned target sub-tasks before executing the above-mentioned target sub-tasks through the P above-mentioned target sub-network models and in accordance with the target execution order, wherein the above-mentioned K is a natural number greater than or equal to the above-mentioned N; a fourth determination module, which is used to determine the target dependency relationship between the N above-mentioned target sub-tasks according to the K above-mentioned target sub-functions, wherein the above-mentioned target dependency relationship is used to represent the various steps of executing the N above-mentioned target tasks and the association relationship between each of the above-mentioned steps; a fifth determination module, which is used to determine the initial execution order of executing the N above-mentioned target sub-tasks from the above-mentioned target dependency relationship; a first adjustment module, which is used to adjust the above-mentioned initial execution order based on the above-mentioned historical execution results and the N groups of the above-mentioned sub-attribute information to obtain the above-mentioned target execution order.

[0015] In an exemplary embodiment, the first screening module comprises: a first execution submodule, used to input the test set into the M sub-network models, so as to perform test operations on the test set through the M sub-network models to obtain the M test results; a first calculation submodule, used to calculate the performance indicators of the M sub-network models through the matching results between the test results and the preset test results, wherein the preset test results are pre-set results based on the functions implemented by the test set; a second screening submodule, used to screen out P target sub-network models from the M sub-network models according to the performance indicators and the historical execution results, wherein the target sub-network model is a model in which the performance indicators of multiple sub-network models meet the preset screening conditions and the historical execution results meet the preset execution results.

[0016] In an exemplary embodiment, the fourth determination module includes: a first determination submodule, used to determine the target information for implementing the K target subfunctions mentioned above, wherein the target information includes input and output information, resource information, time information and priority information when executing the target subtasks, the input and output information is used to determine the input and output dependencies for executing the N target subtasks mentioned above, the resource information is used to determine the resource dependencies for executing the N target subtasks mentioned above, and the time information and the priority information are used to determine the execution dependencies for executing the N target subtasks mentioned above; a second determination submodule, used to determine the input and output dependencies, the resource dependencies and the execution dependencies as the target dependencies.

[0017] In an exemplary embodiment, the above-mentioned device also includes: a first collection module, which is used to collect feedback information when the above-mentioned target sub-network model executes the above-mentioned target sub-tasks during the execution process, after executing the above-mentioned N target sub-tasks in the target execution order through P of the above-mentioned target sub-network models, wherein the above-mentioned feedback information is used to indicate the execution status of the above-mentioned target sub-tasks; a second collection module, which is used to collect result information when the above-mentioned target sub-network model executes the above-mentioned target sub-tasks after the execution is completed; and a third determination sub-module, which is used to determine the above-mentioned feedback information and the above-mentioned result information as the target execution result.

[0018] According to another embodiment of the present application, a computer program product is provided, including a computer program, wherein the computer program is configured for a processor to execute the steps of any of the above method embodiments.

[0019] According to another embodiment of the present application, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to enable a processor to execute the steps in any of the above method embodiments.

[0020] According to another embodiment of the present application, an electronic device is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor is configured to execute the computer program to perform the steps in any one of the method embodiments.

[0021] Through this application, the target task is split based on the attribute information to obtain the target subtask. According to the subtask type of the target subtask, the subnetwork model that can execute the target subtask is determined. The target subnetwork model is screened out from the subnetwork model based on the test results obtained by executing the test set through the subnetwork model. Finally, based on the execution order determined by the attribute information and historical execution results, N target subtasks are executed. Since this application can automatically identify the attribute information of the target task, complex tasks can be split into smaller and more specific target subtasks. According to the type of each target subtask and the performance of the subnetwork model at the target subtask type, the target subnetwork model that is most suitable for executing this type of target subtask is selected from the subnetwork model. Finally, according to the attribute information and historical execution results, the execution order of the target subtask is dynamically determined. Therefore, the problem of the inability to effectively execute tasks in the related art is solved, and the effect of effectively executing tasks is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a hardware structure block diagram of a mobile terminal of a method for executing a task in an embodiment of the present application;

[0023] Figure 2 is a flowchart of a method for executing a task according to an embodiment of the present application;

[0024] Figure 3 It is a flowchart of a method for executing a task in a specific embodiment of the present application;

[0025] Figure 4 It is an architecture diagram of a method for executing a task in a specific embodiment of the present application;

[0026] Figure 5 It is a structural block diagram of a task execution device according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] The embodiments of the present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0029] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Taking running on a mobile terminal as an example, Figure 1 1 is a hardware structure block diagram of a mobile terminal for executing a task in an embodiment of the present application. Figure 1 As shown, the mobile terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the mobile terminal may also include a transmission device 106 and an input / output device 108 for communication functions. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the mobile terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0030] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as a computer program corresponding to a method for executing a task in an embodiment of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, implementing the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the mobile terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0031] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0032] In this embodiment, a method for executing a task is provided. Figure 2 is a flowchart of a method for executing a task according to an embodiment of the present application, such as Figure 2As shown, the process includes the following steps:

[0033] Step S202, identifying attribute information of the target task, and splitting the target task based on the attribute information to obtain N target subtasks, where N is a natural number greater than or equal to 1;

[0034] Optionally, the target task includes but is not limited to handling customer complaints, recommending products, evaluating customer credit, and other tasks that can be handled by a large language model.

[0035] Optionally, the attribute information is used to represent requirement information for achieving the target task, such as input type, output type, processing power requirement, and time limit.

[0036] Step S204, determining M sub-network models according to the sub-task types of the N target sub-tasks, wherein the M is a natural number greater than or equal to 1, the sub-network models are used to perform the target sub-tasks, and one sub-task type corresponds to at least one sub-network model;

[0037] Optionally, the subtask type includes but is not limited to text recognition type, speech recognition type, and image processing type. The subnetwork model includes but is not limited to a base model for text language analysis or generation, a vertical model for image generation or video generation, and a traditional discriminant model for speech recognition.

[0038] Step S206, obtaining test results by executing the test set through the M sub-network models, and screening out P target sub-network models from the M sub-network models according to the test results, wherein P is a natural number less than or equal to N, one sub-task type corresponds to one test set, and one test set corresponds to at least one sub-network model;

[0039] Step S208, executing the N target subtasks through the P target sub-network models and in a target execution order, wherein the target execution order is an execution order determined based on the attribute information and historical execution results.

[0040] Optionally, through modular design and standardized interfaces, target sub-network models can be updated and replaced independently.

[0041] The executor of the above steps in this embodiment can be a terminal, a server, a specific processor set in the terminal or server, or a processor or processing device set relatively independently from the terminal or server, but is not limited to this. For example, a system that deploys the big model as a service technology.

[0042] Alternatively, for example, in an online shopping platform, customers may request a return due to product quality problems, inappropriate size, or product description discrepancy. The platform needs to provide efficient and fair return processing services based on factors such as inventory, sales impact, and customer satisfaction. First, the large language model uses the language recognition capability of its own base model to identify the attribute information of the return processing service: Input type: customer return request (including order information, product details, return reason description), customer historical shopping record, platform return policy, current inventory status, and cost data. Output type: return qualification confirmation (whether return is allowed), return reason classification, return impact assessment, return guidance document, subsequent operation instructions (logistics scheduling, financial refund); text understanding ability, parsing customer return reason description; decision analysis ability, assessing return impact and formulating strategies. Based on the above attribute information, the return processing service is divided into 4 target subtasks: verifying return qualification, confirming return reason, assessing return impact, and generating return guidance. Then, based on the subtask types of verifying return eligibility, confirming return reasons, evaluating return impact, and generating return guidance, the logistic regression subnetwork model and decision tree subnetwork model for verifying return eligibility are determined, the bidirectional encoder representation (BERT) subnetwork model and text-to-text transfer transformer (T5) subnetwork model for confirming return reasons and generating return guidance are determined, and the linear regression subnetwork model and neural network model subnetwork model for evaluating return impact are determined. The test results obtained by executing the test set by the above subnetwork models are obtained, and the target subnetwork model is screened out from the above subnetwork models according to the test results: the logistic regression subnetwork model for verifying return eligibility, the BERT subnetwork model for confirming return reasons and generating return guidance, and the neural network model subnetwork model for evaluating return impact. Through the above target subnetwork model and in the target execution order of verifying return eligibility → confirming return reasons → generating return guidance → evaluating return impact, N above target subtasks are executed.

[0043] Through the above steps, the target task is split based on the attribute information to obtain the target subtask. According to the subtask type of the target subtask, the subnetwork model that can execute the target subtask is determined. The target subnetwork model is screened out from the subnetwork model based on the test results obtained by executing the test set with the subnetwork model. Finally, based on the execution order determined by the attribute information and historical execution results, N target subtasks are executed. Since the present application can automatically identify the attribute information of the target task, complex tasks can be split into smaller and more specific target subtasks. According to the type of each target subtask and the performance of the subnetwork model at the target subtask type, the target subnetwork model that is most suitable for executing this type of target subtask is selected from the subnetwork model. Finally, according to the attribute information and historical execution results, the execution order of the target subtasks is dynamically determined. Therefore, the problem of the inability to effectively execute tasks in the related art is solved, and the effect of effectively executing tasks is achieved.

[0044] In an exemplary embodiment, before identifying the attribute information of the target task and splitting the target task based on the attribute information to obtain N target subtasks, the method further includes: acquiring task information and context information of the target task, wherein the task information is used to represent the function implemented by the target task, and the context information is used to represent the situational background of the target task; and determining the attribute information of the target task from the task information and the context information.

[0045] Optionally, the context information includes, but is not limited to, background information, prior information, domain knowledge, or specific situational information related to the task, question, or dialogue. Optionally, obtaining the task information and context information of the target task usually requires collection and integration from multiple sources and channels. For example, the large language model obtains the task information and context information of the target task by analyzing business requirement documents, customer feedback and complaints, historical data and logs, industry standards, and regulations.

[0046] Optionally, the target task is to handle customer complaints as an example: the task information includes but is not limited to: automatically analyzing and understanding the content of customer complaints, identifying specific problems such as product quality, logistics delays or customer service attitude, etc.; based on the identified problems, generating corresponding solutions or compensation suggestions, such as refunds, exchanges, coupon compensation, etc.; evaluating the impact of customer complaint handling on the company's reputation, customer relationships and operating costs. Context information includes: the e-commerce platform receives a large number of customer complaints during a specific promotion period, mainly involving product defects and logistics problems; it is necessary to obtain relevant information from warehouses, logistics, finance and customer service departments; it is necessary to use big data analysis to optimize inventory management, logistics scheduling and customer service, and complaint data is important feedback information. Generate attribute information based on the above task information and context information: Input type: specific description of customer complaints, order and product information, company service policies, historical complaint records and processing results. Output type: classification of complaint issues, solutions or compensation suggestions, impact assessment reports, processing progress notifications, process improvement suggestions. Resource requirements: database access rights, including customer information, order information and historical complaint records; and interface rights with external services, such as logistics and payment services.

[0047] This embodiment achieves the purpose of significantly improving the design accuracy and execution efficiency of the target task by acquiring and analyzing the task information and context information of the target task, and then determining the attribute information and performing task splitting.

[0048] In an exemplary embodiment, the attribute information of the target task is identified to split the target task based on the attribute information to obtain N target subtasks, including: splitting the functions implemented by the target task based on the classification of the attribute information, and combining the functions implemented by the split target tasks according to the functions implemented by the target task to obtain multiple function combinations; screening out a target function combination from the multiple function combinations based on the historical execution results, wherein the target function combination is a function combination whose historical execution results meet the preset execution results among the multiple function combinations; splitting the target task according to the target function combination to obtain N target subtasks.

[0049] Optionally, the target task is to optimize and upgrade the intelligent customer service of an e-commerce platform. For example, the attribute information includes: Input type: customer query, complaint text, product information, historical conversation record. Output type: automatic reply, solution suggestion, transfer to manual service instruction, product recommendation, etc. Processing capability requirement: natural language processing capability, sentiment analysis, decision tree or machine learning model. Classification splitting: Function A: Automatic reply generation - Generate instant replies based on customer queries. Function B: Sentiment analysis and emotion regulation - Analyze customer emotions and adjust the tone and content of the reply. Function C: Problem solution suggestion - Provide solution strategies for customer complaints or problems. Function D: Transfer to manual service judgment - Identify complex problems and judge whether they need to be transferred to manual customer service. Function E: Product recommendation and guidance - Recommend related products based on customer conversations and purchase history. Function combination: (A+E) and (B+C+D) Pre-sales consultation combination (A+E): When a customer inquires about a product or service, the system can quickly generate a reply with relevant product information and recommend products that may be of interest to the customer based on the content of the customer conversation. After-sales problem handling combination (B+C+D): When customers have after-sales problems or complaints, the system first analyzes emotions and responds with a warmer and more understanding tone, then provides targeted solution suggestions. If it cannot be solved automatically, it will automatically judge and transfer to manual customer service. Full-cycle intelligent customer service combination (A+B+C+D+E): Integrate pre-sales, in-sales, and after-sales customer service functions to build an intelligent customer service that can understand customer needs, provide personalized services, quickly solve problems, transfer to manual when necessary, and recommend products. By analyzing historical execution results, it can be determined which function combination can best improve customer satisfaction and business efficiency. For example, if the "full-cycle intelligent customer service combination" performs well in handling after-sales problems, can significantly reduce customer waiting time and improve satisfaction, then this combination can be used as the target function combination. Target subtask split: Target subtask 1: Quickly generate automatic replies for different inquiries and needs before, during, and after sales. Target subtask 2: Analyze customer emotions and adjust the response strategy and tone to better respond to customer emotions. Target subtask 3: Generate problem solution suggestions, when customers raise questions or complaints, provide solutions or remedial measures. Target subtask 4: Automatically determine when to transfer complex problems to manual customer service. Target subtask 5: Personalized product recommendation service.

[0050] This embodiment decomposes the complex target task into N more specific and smaller target subtasks, each of which can be optimized independently, thereby achieving the purpose of fine management of the target task and reducing the complexity of the target task. At the same time, by screening the target function combination that meets the preset execution result of the historical execution results, the strategy can be adjusted according to the previous operation data and feedback, and the optimal function combination can be selected, thereby achieving the purpose of optimizing the execution of the current target task.

[0051] In an exemplary embodiment, before executing the N target subtasks through the P target sub-network models and in the target execution order, the method further includes: respectively determining the K target sub-functions implemented by the N target subtasks and the N groups of sub-attribute information corresponding to the N target subtasks, wherein the K is a natural number greater than or equal to the N; determining the target dependency relationship between the N target subtasks according to the K target subfunctions, wherein the target dependency relationship is used to represent the individual steps of executing the N target tasks and the association relationship between the individual steps; determining the initial execution order of executing the N target subtasks from the target dependency relationship; adjusting the initial execution order based on the historical execution results and the N groups of sub-attribute information to obtain the target execution order.

[0052] Optionally, the target execution order may be the initial execution order. The sub-attribute information is used to indicate the requirement information for achieving the target sub-task, for example, input type, output type, processing capacity requirement, and time limit.

[0053] Optionally, the target task is to optimize and upgrade the intelligent customer service of the e-commerce platform as an example. For example, target subtask 1: quickly generate automatic replies for different inquiries and needs before, during and after sales, target subtask 2: analyze customer emotions, adjust the reply strategy and tone to better respond to customer emotions. Target subtask 3: generate problem solution suggestions, and provide solutions or remedial measures when customers raise questions or complaints. Target subtask 4: automatically determine when to transfer complex problems to manual customer service. Target subtask 5: personalized product recommendation service. For each target subtask, further determine the target subfunctions to be achieved and the corresponding sub-attribute information. For example, the target subfunctions that target subtask 1 needs to achieve include: natural language understanding, query intent recognition, and keyword extraction; the sub-attribute information may include: input type: text, output type: intent recognition results, keyword list. Next, analyze the dependencies between these target subtasks to build a reasonable execution process. For example, customer query parsing (target subtask 1) is the prerequisite for sentiment analysis (target subtask 2), and the result of sentiment analysis affects the way problem solution is generated (target subtask 3), that is: target subtask 1 → target subtask 2 → target subtask 3.

[0054] Optionally, the execution order can be optimized by analyzing historical execution results and the sub-attribute information of each target subtask. For example, historical execution results show that most customers will directly request product recommendations after asking for product information, and sentiment analysis can improve the efficiency of problem solution generation in most cases. In this case, product recommendations can be advanced to before problem solution generation to respond to customers' direct needs more quickly.

[0055] This embodiment achieves the purpose of building a logically coherent and clearly structured execution process by determining target dependencies. At the same time, it adjusts the initial execution order based on historical execution results and sub-attribute information, and can automatically select the optimal execution path according to real-time conditions and historical data, thereby achieving the purpose of improving the flexibility and adaptability of task execution.

[0056] In an exemplary embodiment, the test results obtained by executing the test set through the M above-mentioned sub-network models, and P target sub-network models are screened out from the M above-mentioned sub-network models according to the above-mentioned test results, including: inputting the above-mentioned test set into the M above-mentioned sub-network models, so as to perform the test operation on the above-mentioned test set through the M above-mentioned sub-network models, and obtain the M above-mentioned test results; calculating the performance indicators of the M above-mentioned sub-network models through the matching results between the above-mentioned test results and the preset test results, wherein the above-mentioned preset test results are pre-set results based on the functions implemented by the above-mentioned test set; according to the above-mentioned performance indicators and the above-mentioned historical execution results, screening out P above-mentioned target sub-network models from the M above-mentioned sub-network models, wherein the above-mentioned target sub-network model is a model in which the above-mentioned performance indicators meet the preset screening conditions among multiple sub-network models, and the above-mentioned historical execution results meet the preset execution results.

[0057] Optionally, the performance indicators include but are not limited to accuracy, precision, recall, and F1 score. The selection of performance indicators can be adjusted based on the requirements of the target subtask. For example, a classification task may focus on accuracy, while a regression task may focus more on mean square error or root mean square error. Optionally, the preset execution results include but are not limited to actual operation indicators such as response time and error rate of the subnetwork model.

[0058] Optionally, one subtask type corresponds to one target subnetwork model.

[0059] Optionally, for example, the target task is to process a large number of product images and identify product types. The test set includes various types of e-commerce product images, and each image has corresponding labels, such as product categories (such as clothing, electronic products, books, etc.), colors, brands, etc. For each image in the test set, the preset test results are set based on the classification and recommendation results of the image by professional product identification personnel, and these results represent the expected accuracy and relevance levels. The M sub-network models include language models: sub-network model 1, sub-network model 2, and image recognition models: sub-network model 3, sub-network model 4, and sub-network model 5. The language test set in the test set is input into sub-network model 1 and sub-network model 2 respectively, and the image recognition test set in the test set is input into sub-network model 3, sub-network model 4, and sub-network model 5 respectively. Each sub-network model analyzes and identifies the input test set, and outputs the classification results and the recommended product list. The output results of each sub-network model are compared with the preset test results, and key performance indicators such as accuracy, recall rate, F1 score, and recommended relevance score are calculated. For example, accuracy can measure the model's ability to correctly classify products, while relevance scores reflect the degree of match between recommended products and customer interests. Based on the performance indicators of the current test set and historical execution results, the language model: sub-network model 1 and the image recognition model: sub-network model 3 are selected from the above sub-network models, whose response time and error rate are less than the preset execution results, and whose accuracy and relevance scores meet the preset screening conditions.

[0060] This embodiment can ensure that the selected model meets the requirements in terms of accuracy, efficiency, stability, etc. by calculating the model performance index and comparing it with the preset standards. At the same time, the screening process not only considers the performance of the model on the test set, but also combines the historical execution results, which helps to reasonably allocate computing resources and avoids wasting resources on models with poor performance or instability, thus achieving the purpose of ensuring that the performance of the sub-network model is fully evaluated during the screening process.

[0061] In an exemplary embodiment, determining the target dependency between the N target subtasks mentioned above according to the K target subfunctions mentioned above includes: determining the target information for realizing the K target subfunctions mentioned above, wherein the target information includes input and output information, resource information, time information and priority information when executing the target subtasks, the input and output information is used to determine the input and output dependencies for executing the N target subtasks mentioned above, the resource information is used to determine the resource dependencies for executing the N target subtasks mentioned above, and the time information and the priority information are used to determine the execution dependencies for executing the N target subtasks mentioned above; and determining the input and output dependencies, the resource dependencies and the execution dependencies as the target dependencies.

[0062] Optionally, the target information can be determined based on the attribute information. Resource information includes but is not limited to model resources, data resources and tool resources. Time information includes but is not limited to the time of responding to the target subtask and the data update time. Priority information includes but is not limited to: the priority of executing the task.

[0063] It can be explained by taking the recognition of customer emotions as an example. Determine the target information of the target sub-function: Get customer information (target sub-task 1): Input and output information: Input is customer registration information, and output is customer voice and customer text. Resource information: It is necessary to call the language understanding model and retrieval module. Time information: It is required to return the answer within 5 seconds. Priority information: High priority.

[0064] Speech emotion recognition (target subtask 2): Input and output information: The input is customer speech, and the output is customer emotion resource information 1. Resource information: It is necessary to call the text generation model and formatting tool. Time information: It is required to return the answer within 10 seconds after the input is the customer speech. Priority information: Medium priority.

[0065] Text emotion recognition (target subtask 3): Input and output information: Input is customer text, output is customer emotion resource information 2. Resource information: Need to call text generation model and formatting tool. Time information: Requires to return the answer within 10 seconds after the input is customer text. Priority information: Medium priority.

[0066] Emotion recognition determination (target subtask 4): Input and output information: The input is customer emotion resource information 1 and customer emotion resource information 2, and the output is customer emotion. Resource information: It is necessary to call the text generation model and formatting tool. Time information: It is required to return the answer within 10 seconds after inputting customer emotion resource information 1 and customer emotion resource information 2. Priority information: High priority.

[0067] Input-output dependency: The input of target subtask 4 depends on the output of target subtask 2 and target subtask 3, and the input of target subtask 2 and target subtask 3 depends on the output of target subtask 1. Resource dependency: There is no model conflict or tool conflict in the execution of target subtasks, so the resource dependency is not dependent. Execution dependency: Execute target subtask 1 first, then execute target subtask 2 and target subtask 3 synchronously, and finally execute target subtask 4. Or execute target subtask 1, target subtask 2, target subtask 3, and target subtask 4 in sequence.

[0068] This embodiment achieves the purpose of being able to flexibly handle various tasks and ensure that high-priority tasks are executed first, while reasonably allocating resources and avoiding resource conflicts and waste by determining target dependencies.

[0069] In an exemplary embodiment, after executing N of the above-mentioned target subtasks through P of the above-mentioned target sub-network models and in a target execution order, the above-mentioned method further includes: collecting feedback information when the above-mentioned target sub-network model executes the above-mentioned target subtask during the execution process, wherein the above-mentioned feedback information is used to indicate the execution status of the above-mentioned target subtask; collecting result information when the above-mentioned target sub-network model executes the above-mentioned target subtask after the execution is completed; and determining the above-mentioned feedback information and the above-mentioned result information as the target execution result.

[0070] Optionally, the feedback information includes, but is not limited to, response time information, error and exception information, and task execution flow feedback when the target sub-network model executes the target sub-task. The result information includes, but is not limited to, performance indicator information, execution time information, and resource consumption information. For example, response time information: When a customer asks the question "How do recent market trends affect our company's sales strategy?", the time from receiving the question to generating a preliminary answer is recorded, for example, the response time is 20 seconds. If the response time exceeds expectations (such as more than 10 seconds), this will serve as feedback information, indicating the need to optimize the performance of the model or increase computing resources.

[0071] The present invention will be described below in conjunction with specific embodiments:

[0072] This embodiment takes the consumer finance scenario work order incoming complaint customer as an example to illustrate the target task: to determine the subsequent processing method for the customer, where the big model as a service technology is deployed in the business processing system. Figure 3 It is a flowchart of a method for executing a task in a specific embodiment of the present application, comprising the following steps:

[0073] S302, such as Figure 4 As shown, Figure 4This is an architecture diagram of a task execution method of a specific embodiment of the present application. When the business layer initiates a target task request to the management layer, the base model analyzes the target task and arranges the overall plan. The base model first obtains the task information and context information of the target task: the task information includes but is not limited to: automatically analyzing and understanding the content of customer complaints, identifying specific problems: customer service attitude, etc.; based on the identified problems, generating corresponding solutions or compensation suggestions, such as refunds, exchanges, coupon compensation, etc. The context information includes: reading the customer's recent process records. Generate attribute information based on the above task information and context information: Input type: specific description of customer complaints. Output type: classification of complaint problems, solutions or compensation suggestions. Resource requirements: customer information and communication data, etc. The base model disassembles and arranges the target task based on language knowledge as follows: Target subtask 1: Read customer user information and portrait tags from the customer registration information user center system. Target subtask 2: Based on user information, obtain user classification scores. Target subtask 3: Identify the voice call records of customer inbound voice manual customer service. Target subtask 4: Analyze customer emotional levels. Target subtask 5: Analyze the reasons that lead to customer complaints. Target subtask 6: Get a compensation plan. Target subtask 7: Implement the compensation plan. Initial execution order: (1→2→3→4→5→6→7).

[0074] S304, according to the subtask type of the above target subtask, determine that the target subtask 1 can be executed: sub-network model A and sub-network model B. The target subtask 2 can be executed: sub-network model E. The target subtask 3 can be executed: sub-network model A and sub-network model B. The target subtask 4 can be executed: sub-network model A and sub-network model B. The target subtask 5 can be executed: sub-network model A and sub-network model B. The target subtask 6 can be executed: sub-network model A and sub-network model B. The target subtask 7 can be executed: sub-network model F in the external tool.

[0075] S306, determine the target sub-network model based on the test results obtained by performing the test operation on the test set through the sub-network model: Target sub-task 1: sub-network model A. Target sub-task 2: sub-network model E. Target sub-task 3: sub-network model B. Target sub-task 4: sub-network model B. Target sub-task 5: sub-network model B. Target sub-task 6: sub-network model A. Target sub-task 7: sub-network model F in the external tool; adjust the initial execution order based on the historical execution results and the sub-attribute information corresponding to the target sub-task, and obtain the target execution order is the same as the initial execution order: (1→2→3→4→5→6→7).

[0076] S308, standardize and adapt the interfaces of each model, and use the target sub-network model to perform the target task. Update the model memory as needed, such as updating the model "knowledge" through prompt word engineering, external vector knowledge base, or parameter fine-tuning. The interfaces of each model need to be standardized and adapted.

[0077] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods of each embodiment of the present application.

[0078] In the present embodiment, a task execution device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware is also possible and conceived.

[0079] Figure 5 is a structural block diagram of a task execution device according to an embodiment of the present application, such as Figure 5 As shown, the device comprises:

[0080] A first identification module 502 is used to identify attribute information of a target task, so as to split the target task based on the attribute information to obtain N target subtasks, where N is a natural number greater than or equal to 1;

[0081] A first determining module 504 is used to determine M sub-network models according to the sub-task types of the N target sub-tasks, wherein the M is a natural number greater than or equal to 1, the sub-network models are used to perform the target sub-tasks, and one sub-task type corresponds to at least one sub-network model;

[0082] A first screening module 506 is used to obtain test results by executing the test set through the M sub-network models, and screen out P target sub-network models from the M sub-network models according to the test results, wherein P is a natural number less than or equal to N, one sub-task type corresponds to one test set, and one test set corresponds to at least one sub-network model;

[0083] The first execution module 508 is used to execute the N target subtasks through the P target sub-network models and in a target execution order, wherein the target execution order is an execution order determined based on the attribute information and historical execution results.

[0084] In an exemplary embodiment, the above-mentioned device also includes: a first acquisition module, used to identify the attribute information of the target task, so as to split the above-mentioned target task based on the above-mentioned attribute information, and before obtaining N target subtasks, acquire the task information and context information of the above-mentioned target task, wherein the above-mentioned task information is used to represent the function implemented by the above-mentioned target task, and the above-mentioned context information is used to represent the situational background of the above-mentioned target task; a second determination module, used to determine the attribute information of the above-mentioned target task from the above-mentioned task information and the above-mentioned context information.

[0085] In an exemplary embodiment, the above-mentioned first identification module includes: a first splitting sub-module, which is used to split the functions implemented by the above-mentioned target tasks based on the classification of the above-mentioned attribute information, and split the functions implemented by the above-mentioned target tasks according to the function combinations implemented by the above-mentioned target tasks, to obtain multiple function combinations; a first screening sub-module, which is used to screen out a target function combination from the multiple function combinations based on the above-mentioned historical execution results, wherein the above-mentioned target function combination is a function combination whose historical execution results meet the preset execution results among the multiple function combinations; a second splitting sub-module, which is used to split the above-mentioned target task according to the above-mentioned target function combination, to obtain N of the above-mentioned target sub-tasks.

[0086] In an exemplary embodiment, the above-mentioned device also includes: a third determination module, which is used to determine the K target sub-functions implemented by the N above-mentioned target sub-tasks and the N groups of sub-attribute information corresponding to the N above-mentioned target sub-tasks before executing the above-mentioned target sub-tasks through the P above-mentioned target sub-network models and in accordance with the target execution order, wherein the above-mentioned K is a natural number greater than or equal to the above-mentioned N; a fourth determination module, which is used to determine the target dependency relationship between the N above-mentioned target sub-tasks according to the K above-mentioned target sub-functions, wherein the above-mentioned target dependency relationship is used to represent the various steps of executing the N above-mentioned target tasks and the association relationship between each of the above-mentioned steps; a fifth determination module, which is used to determine the initial execution order of executing the N above-mentioned target sub-tasks from the above-mentioned target dependency relationship; a first adjustment module, which is used to adjust the above-mentioned initial execution order based on the above-mentioned historical execution results and the N groups of the above-mentioned sub-attribute information to obtain the above-mentioned target execution order.

[0087] In an exemplary embodiment, the first screening module comprises: a first execution submodule, used to input the test set into the M sub-network models, so as to perform test operations on the test set through the M sub-network models to obtain the M test results; a first calculation submodule, used to calculate the performance indicators of the M sub-network models through the matching results between the test results and the preset test results, wherein the preset test results are pre-set results based on the functions implemented by the test set; a second screening submodule, used to screen out P target sub-network models from the M sub-network models according to the performance indicators and the historical execution results, wherein the target sub-network model is a model in which the performance indicators of multiple sub-network models meet the preset screening conditions and the historical execution results meet the preset execution results.

[0088] In an exemplary embodiment, the fourth determination module includes: a first determination submodule, used to determine the target information for implementing the K target subfunctions mentioned above, wherein the target information includes input and output information, resource information, time information and priority information when executing the target subtasks, the input and output information is used to determine the input and output dependencies for executing the N target subtasks mentioned above, the resource information is used to determine the resource dependencies for executing the N target subtasks mentioned above, and the time information and the priority information are used to determine the execution dependencies for executing the N target subtasks mentioned above; a second determination submodule, used to determine the input and output dependencies, the resource dependencies and the execution dependencies as the target dependencies.

[0089] In an exemplary embodiment, the above-mentioned device also includes: a first collection module, which is used to collect feedback information when the above-mentioned target sub-network model executes the above-mentioned target sub-tasks during the execution process, after executing the above-mentioned N target sub-tasks in the target execution order through P of the above-mentioned target sub-network models, wherein the above-mentioned feedback information is used to indicate the execution status of the above-mentioned target sub-tasks; a second collection module, which is used to collect result information when the above-mentioned target sub-network model executes the above-mentioned target sub-tasks after the execution is completed; and a third determination sub-module, which is used to determine the above-mentioned feedback information and the above-mentioned result information as the target execution result.

[0090] An embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps in any one of the above method embodiments are implemented.

[0091] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0092] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0093] An embodiment of the present application further provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0094] In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0095] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0096] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0097] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for executing a task, characterized in that: include: Identify attribute information of a target task, and split the target task based on the attribute information to obtain N target subtasks, where N is a natural number greater than or equal to 1; Determine M sub-network models according to the sub-task types of the N target sub-tasks, wherein M is a natural number greater than or equal to 1, the sub-network model is used to perform the target sub-task, and one sub-task type corresponds to at least one sub-network model; The test results obtained by executing the test set by the M sub-network models, and P target sub-network models are screened out from the M sub-network models according to the test results, wherein P is a natural number less than or equal to N, one sub-task type corresponds to one test set, and one test set corresponds to at least one sub-network model; The N target subtasks are executed through the P target subnetwork models and in a target execution order, wherein the target execution order is an execution order determined based on the attribute information and historical execution results.

2. The method according to claim 1, characterized in that Before identifying attribute information of the target task and splitting the target task based on the attribute information to obtain N target subtasks, the method further includes: Acquire task information and context information of the target task, wherein the task information is used to indicate the function implemented by the target task, and the context information is used to indicate the scenario background of the target task; Attribute information of the target task is determined from the task information and the context information.

3. The method according to claim 1, characterized in that Identify attribute information of the target task, and split the target task based on the attribute information to obtain N target subtasks, including: Splitting the functions realized by the target tasks based on the classification of the attribute information, and combining the split functions realized by the target tasks according to the functions realized by the target tasks to obtain a plurality of function combinations; Filtering out a target function combination from the plurality of function combinations based on the historical execution results, wherein the target function combination is a function combination whose historical execution results satisfy a preset execution result among the plurality of function combinations; The target task is split according to the target function combination to obtain N target subtasks.

4. The method according to claim 1, characterized in that: Before executing the N target subtasks in a target execution order by using the P target subnetwork models, the method further includes: Respectively determine K target sub-functions implemented by the N target sub-tasks and N groups of sub-attribute information corresponding to the N target sub-tasks, wherein K is a natural number greater than or equal to N; Determine a target dependency relationship between the N target subtasks according to the K target subfunctions, wherein the target dependency relationship is used to represent the steps of executing the N target tasks and the association relationship between the steps; Determining an initial execution order of executing the N target subtasks from the target dependency relationship; The initial execution order is adjusted based on the historical execution results and the N groups of sub-attribute information to obtain the target execution order.

5. The method according to claim 1, characterized in that The test results obtained by executing the test set by the M sub-network models, and P target sub-network models are screened out from the M sub-network models according to the test results, including: Inputting the test set into the M sub-network models, so as to perform a test operation on the test set through the M sub-network models, and obtain M test results; Calculating the performance indicators of the M sub-network models through the matching results between the test results and the preset test results, wherein the preset test results are pre-set results based on the functions implemented by the test set; According to the performance indicators and the historical execution results, P target sub-network models are screened out from the M sub-network models, wherein the target sub-network model is a model whose performance indicators among the multiple sub-network models meet the preset screening conditions and whose historical execution results meet the preset execution results.

6. The method according to claim 4, characterized in that Determining the target dependency relationship between the N target subtasks according to the K target subfunctions includes: Determine target information for implementing K target sub-functions, wherein the target information includes input and output information, resource information, time information and priority information when executing the target subtasks, the input and output information is used to determine the input and output dependencies of executing N target subtasks, the resource information is used to determine the resource dependencies of executing N target subtasks, and the time information and the priority information are used to determine the execution dependencies of executing N target subtasks; The input-output dependency, the resource dependency, and the execution dependency are determined as the target dependency.

7. The method according to claim 1, characterized in that After executing the N target subtasks through the P target subnetwork models and in the target execution order, the method further includes: Collecting feedback information when the target sub-network model executes the target sub-task during the execution process, wherein the feedback information is used to indicate the execution status of the target sub-task; After the collection is completed, the result information of the target sub-network model executing the target sub-task; The feedback information and the result information are determined as the target execution result.

8. A task execution device, characterized in that: include: A first identification module is used to identify attribute information of a target task, so as to split the target task based on the attribute information to obtain N target subtasks, where N is a natural number greater than or equal to 1; A first determination module is used to determine M sub-network models according to the sub-task types of the N target sub-tasks, wherein M is a natural number greater than or equal to 1, the sub-network model is used to perform the target sub-task, and one sub-task type corresponds to at least one sub-network model; A first screening module is used to obtain test results by executing the test set through the M sub-network models, and screen out P target sub-network models from the M sub-network models according to the test results, wherein P is a natural number less than or equal to N, one sub-task type corresponds to one test set, and one test set corresponds to at least one sub-network model; The first execution module is used to execute the N target subtasks through the P target sub-network models and in a target execution order, wherein the target execution order is an execution order determined based on the attribute information and historical execution results.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program implements the steps of the method described in any one of claims 1 to 7 when executed by a processor.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method described in any one of claims 1 to 7 are implemented.

Images