Business processing method, business processing system, vehicle and storage medium

Abstract

The invention provides a business processing method, a business processing system, a vehicle and a storage medium, and the method relates to the field of vehicles, and comprises the steps: obtaining a business consultation request of a user; determining at least two target task units corresponding to the business consultation request; calling a target model in a pre-configuration model through the at least two target task units to obtain target reply content of the business consultation request; and outputting the target reply content. According to the method, the business processing efficiency in the car business scene can be improved.

Description

Technical Field

[0001] The present application relates to the field of vehicles, and more specifically, to a service processing method, a service processing system, a vehicle, and a storage medium in the field of vehicles. Background Art

[0002] By applying big models to automotive business scenarios, the intelligence level and business innovation capabilities of automotive businesses can be improved.

[0003] However, due to the high technical barriers to entry for large models, automotive service providers must independently understand the model structure and inference interface during use, increasing the cost of integrating and learning large models. Furthermore, different automotive service scenarios require their own development of large model services, resulting in low business processing efficiency within these scenarios. Therefore, improving business processing efficiency within these scenarios is a technical challenge that currently needs to be addressed. Summary of the Invention

[0004] The present application provides a business processing method, a business processing system, a vehicle and a storage medium, which can improve the business processing efficiency in business consulting scenarios.

[0005] In a first aspect, a service processing method is provided, the method comprising:

[0006] Obtaining users' business consultation requests;

[0007] Determine at least two target task units corresponding to the business consultation request;

[0008] Calling a target model in a preconfigured model through at least two target task units to obtain target response content for the business consultation request;

[0009] Output the target response content.

[0010] In an embodiment of the present application, after obtaining a business consultation request from a user, the target model in the preset configuration model is called according to at least two target task units corresponding to the business consultation request to obtain the target reply content of the business consultation request. Compared with the prior art in which each car company's business scenario develops a large model service, this solution pre-configures different task units and directly calls the target model in the pre-configured model according to the target task unit corresponding to the business consultation request. Through the unified configuration and management of the model, different business consultation requests can all call the model to generate the reply content corresponding to the business consultation request, that is, the pre-configured model is called by the task unit to generate the reply content corresponding to the business consultation request in different business scenarios. This solves the problem of low business processing efficiency caused by the business consultant developing different models according to different business scenarios in the prior art, thereby improving the business processing efficiency in the business consultation scenario.

[0011] In conjunction with the first aspect, in certain implementations of the first aspect, obtaining target response content for a business consultation request by invoking a target model in a preconfigured model through at least two target task units includes:

[0012] Determine at least two prompt word templates for target task units;

[0013] Generate prompt words based on business consultation requests and prompt word templates;

[0014] The prompt words are input into the target model so that the target model generates target response content according to the prompt words.

[0015] In an embodiment of the present application, prompt words are generated based on a business consultation request and prompt word templates for at least two target task units, and are then input into the target model. Because prompt words are the core medium for user interaction with the model, they play a decisive role in guiding the content, direction, and style of the model's output, directly affecting the quality of the model's generated content. Therefore, generating prompt words based on the prompt word templates for the target task units and the business consultation request ensures that high-quality prompt words can be generated for the business consultation request; thereby improving the output of the target model, that is, improving the quality of the generated target response content.

[0016] In combination with the first aspect and the above implementations, in some implementations of the first aspect, the method further includes:

[0017] Based on the business consultation request, determining whether the business consultation request is associated with a knowledge fragment in the target knowledge base;

[0018] Input the prompt words into the target model so that the target model generates target response content according to the prompt words, including:

[0019] If the business consultation request is associated with a knowledge fragment in the target knowledge base, the knowledge fragment and prompt words associated with the business consultation request are input into the target model so that the target model generates target response content based on the prompt words and knowledge fragment; wherein, the target knowledge base is used to store business knowledge related to business scenarios.

[0020] In the embodiments of the present application, if the business consultation request is associated with the knowledge segments in the target knowledge base, the business consultation request and the knowledge segments are input into the target model to generate target reply content. Since the prior art model relies on the knowledge during model training when generating content, if the relevant knowledge in the business scenario is not injected into the model, the content generated by the model may be inaccurate. Therefore, by storing the business knowledge related to the business scenario in the target knowledge base, when the business consultation request is associated with the knowledge segments in the target knowledge base, the knowledge segments and the prompt words in the target knowledge base are input into the target model, so as to ensure that the target model can generate more accurate target reply content according to the knowledge segments and the prompt words in the target knowledge base.

[0021] With reference to the first aspect and the above implementation manners, in some implementation manners of the first aspect, the method further includes:

[0022] The tasks that can be performed by the pre-configured model are split to obtain a plurality of task units associated with the pre-configured model, wherein different task units are used to perform different tasks.

[0023] The at least two target task units corresponding to the business consultation request are determined, including:

[0024] The business consultation request is analyzed to determine the target task corresponding to the business consultation request.

[0025] The target task unit corresponding to the target task is determined from the plurality of task units.

[0026] In the embodiments of the present application, the tasks that can be performed by the pre-configured model are split to obtain a plurality of task units, and the target task unit is determined from the plurality of task units according to the target task corresponding to the business consultation request. Since each model in the pre-configured model has different processing capabilities and can perform different tasks, the tasks that can be performed by the pre-configured model are split to obtain task units that can perform different tasks, so as to facilitate calling different processing capabilities of the model through the task units, analyzing and decomposing the business consultation request into a plurality of target tasks, so as to determine the target task unit from the plurality of task units according to the target task, and ensure that the target task unit can be called to perform the target task.

[0027] With reference to the first aspect and the above implementation manners, in some implementation manners of the first aspect, the method further includes:

[0028] The business process corresponding to the business consultation request is determined.

[0029] The at least two target task units are sorted based on the business process to obtain a target task chain.

[0030] The target model in the pre-configured model is called through the at least two target task units, including:

[0031] The target models are called sequentially according to the order of at least two target task units in the target task chain.

[0032] In an embodiment of the present application, the target task units are sorted according to the business process corresponding to the business consultation request to obtain a target task chain, and the target model is called in sequence according to the order in the target task chain. Since a business consultation request corresponds to multiple target tasks, if the execution order of the target tasks is wrong, it may result in the inability to generate the target reply content or the quality of the generated target reply content is poor. Therefore, the target model is called in sequence according to the sequence of the target task chain to ensure that the target tasks are executed in sequence, thereby calling the model in sequence according to the correct execution order of the business process to generate the target reply content corresponding to the business consultation request.

[0033] In combination with the first aspect and the above implementations, in some implementations of the first aspect, the method further includes:

[0034] Determine whether the user has business consultation permissions;

[0035] Determine at least two target task units corresponding to the business consultation request, including:

[0036] If the user has business consultation authority, at least two target task units corresponding to the business consultation request are determined.

[0037] In an embodiment of the present application, since at least two target tasks corresponding to the business consulting request are determined when the user has business consulting authority, it is ensured that each call behavior to the model is controllable and traceable, thereby preventing users who do not have business consulting authority from calling the model, thereby improving the security of the business consulting service.

[0038] In combination with the first aspect and the above implementations, in some implementations of the first aspect, the method further includes:

[0039] Obtaining usage parameters of the preconfigured model, where the usage parameters include the calling frequency, response time, failure rate, and text unit consumption of the preconfigured model;

[0040] Perform configuration management on the preconfigured model based on the usage parameters of the preconfigured model.

[0041] In an embodiment of the present application, usage parameters of the preconfigured model are obtained, and the preconfigured model is managed according to the usage parameters (for example, more computing power is allocated to models with higher calling frequency), ensuring that the preconfigured model can be managed and optimized according to the usage parameters during the use of the preconfigured model.

[0042] In a second aspect, a service processing device is provided, the interactive service processing device including:

[0043] The acquisition module is used to obtain the user's business consultation request;

[0044] A processing module, configured to determine at least two target task units corresponding to the business consulting request;

[0045] The target model in the preconfigured model is called through at least two target task units to obtain target response content for the business consultation request; and the target response content is output.

[0046] In combination with the second aspect, in certain implementations of the second aspect, the processing module is specifically used to: determine the prompt word templates of at least two target task units; generate prompt words based on the business consultation request and the prompt word template; and input the prompt words into the target model so that the target model generates target reply content based on the prompt words.

[0047] In combination with the second aspect and the above-mentioned implementation methods, in certain implementation methods of the second aspect, the processing module is specifically used to: based on the business consultation request, determine whether the business consultation request is associated with a knowledge fragment in the target knowledge base; if the business consultation request is associated with a knowledge fragment in the target knowledge base, input the knowledge fragment and prompt words associated with the business consultation request into the target model, so that the target model generates target reply content based on the prompt words and the knowledge fragment.

[0048] In combination with the second aspect and the above-mentioned implementation methods, in some implementation methods of the second aspect, the processing module is specifically used to: split the tasks that can be executed by the pre-configured model to obtain multiple task units associated with the pre-configured model, wherein different task units are used to perform different tasks; parse the business consulting request to determine the target task corresponding to the business consulting request; and determine the target task unit corresponding to the target task from multiple task units.

[0049] In combination with the second aspect and the above-mentioned implementation methods, in some implementation methods of the second aspect, the processing module is also used to: determine the business process corresponding to the business consultation request; based on the business process, sort at least two target task units to obtain a target task chain; and call the target model in sequence according to the order of at least two target task units in the target task chain.

[0050] In combination with the second aspect and the above-mentioned implementation methods, in some implementation methods of the second aspect, the processing module is also used to: determine whether the user has business consultation authority; if the user has business consultation authority, determine at least two target task units corresponding to the business consultation request.

[0051] In combination with the second aspect and the above implementation, in some implementations of the second aspect, the acquisition module is further used to: obtain usage parameters of the preconfigured model, wherein the usage parameters include the call frequency, response time, failure rate, and consumption of text units of the preconfigured model;

[0052] The processing module is further used to: perform configuration management on the preconfigured model based on usage parameters of the preconfigured model.

[0053] In a third aspect, a business processing system is provided, which includes an application layer, an intelligent orchestration layer, and a model layer;

[0054] The application layer is used to obtain users' business consultation requests;

[0055] The intelligent orchestration layer is used to determine at least two target task units corresponding to the business consultation request, and call the target model in the preconfigured model of the model layer through the at least two target task units, so that the target model generates target response content;

[0056] The model layer is used to generate target reply content and send it to the application layer so that the application layer outputs the target reply content.

[0057] In combination with the third aspect, in certain implementations of the third aspect, the intelligent orchestration layer is further used to: determine prompt word templates for at least two target task units; generate prompt words based on the business consultation request and the prompt word template; and input the prompt words into the target model so that the target model generates target response content based on the prompt words.

[0058] In combination with the third aspect and the above-mentioned implementation methods, in some implementation methods of the third aspect, the business processing system also includes a knowledge enhancement layer; the knowledge enhancement layer is used to send the knowledge fragments to the intelligent orchestration layer when the business consultation request is associated with the knowledge fragments in the target knowledge base.

[0059] In combination with the third aspect and the above implementations, in certain implementations of the third aspect, the intelligent orchestration layer is used to split tasks executable by the preconfigured model to obtain multiple task units associated with the preconfigured model, where different task units are used to execute different tasks; parse the business consultation request to determine a target task corresponding to the business consultation request; and determine a target task unit corresponding to the target task from the multiple task units;

[0060] In combination with the third aspect and the above-mentioned implementation methods, in certain implementation methods of the third aspect, the intelligent orchestration layer is used to determine the business process corresponding to the business consultation request; based on the business process, at least two target task units are sorted to obtain a target task chain; and the target model is called in sequence according to the order of at least two target task units in the target task chain.

[0061] In combination with the third aspect and the above-mentioned implementation methods, in some implementation methods of the third aspect, the business processing system also includes a service governance layer; the service governance layer is used to: determine whether the user has business access rights, and when the user has business consultation rights, send the business consultation request to the intelligent orchestration layer.

[0062] In combination with the third aspect and the above-mentioned implementation methods, in some implementation methods of the third aspect, the business processing system also includes a platform operation layer, the application layer is used to obtain the user's business consulting request and send the business consulting request to the service governance layer; the platform operation layer is used to obtain the usage parameters of the pre-configured model, wherein the usage parameters include the call frequency, response time, failure rate and text unit consumption of the pre-configured model; based on the usage parameters of the pre-configured model, the pre-configured model is configured and managed.

[0063] In a fourth aspect, a vehicle is provided, comprising a memory and a processor, wherein the memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, so that the vehicle executes the method in the above-mentioned first aspect or any possible implementation of the first aspect.

[0064] In a fifth aspect, a computer program product is provided, which includes: computer program code, which, when running on a computer, enables the computer to execute the method in the above-mentioned first aspect or any possible implementation of the first aspect.

[0065] In a sixth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed, it implements the method in the above-mentioned first aspect or any possible implementation of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 This is an application scenario diagram of the business interaction method provided in an embodiment of the present application;

[0067] Figure 2 This is a schematic flow chart of a business processing method provided in an embodiment of the present application;

[0068] Figure 3 This is a schematic diagram of the system architecture of a business processing system provided by an embodiment of the present application;

[0069] Figure 4 This is a schematic flow chart of another business processing method provided by an embodiment of the present application;

[0070] Figure 5 This is a schematic diagram of the structure of a service processing device provided in an embodiment of the present application;

[0071] Figure 6 It is a structural schematic diagram of a vehicle provided in an embodiment of the present application. DETAILED DESCRIPTION

[0072] The following will clearly and thoroughly describe the technical solutions in this application in conjunction with the accompanying drawings. In the description of the embodiments of this application, unless otherwise specified, " / " means or, for example, A / B can mean A or B: "and / or" in the text is only a description of the association relationship of associated objects, indicating that there can be three relationships, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more than two.

[0073] In the following, the terms "first" and "second" are used for descriptive purposes only and should not be understood to imply or suggest relative importance or implicitly indicate the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features.

[0074] Applying big models to automotive business scenarios can enhance the intelligence and innovation capabilities of these businesses. However, the underlying technical barriers to entry for big models are high, requiring automotive service providers to independently understand the model structure and inference interface during use, increasing the cost of integrating and learning big models. Furthermore, different automotive business scenarios often develop their own big model services, lacking unified configuration and management of the models.

[0075] It should be noted that due to the high degree of expertise inherent in the underlying technologies of large models (e.g., Transformer model architecture, inference engines, and prompt word optimization), business departments (e.g., in-vehicle system developers and smart cockpit designers) need to understand the model structure, inference interface call methods (such as interface parameter configuration), and prompt word design techniques to better integrate large models into automotive business. However, this approach results in high learning costs for business personnel and high business costs associated with integrating large models into automotive business. Furthermore, for different businesses in different automotive business scenarios (e.g., intelligent driving decision-making, connected car user operations, and after-sales customer service), if each business party independently develops large model services without establishing unified standards, unified model configuration and management will be impossible. For example, a business party for the intelligent driving business scenario develops a "road condition prediction large model," while a business party for the connected car scenario develops a "user interest recommendation large model." Both models may use the same Large Language Model (LLM) foundation, but each encapsulates its own interfaces and training data, resulting in duplicated computing resources.

[0076] It can be seen that when the existing technology applies large models to automotive business scenarios, the business processing efficiency of automotive business scenarios is low; therefore, how to improve the business processing efficiency in automotive business scenarios is a technical problem that needs to be solved at present.

[0077] In view of this, the present application provides a business processing method, a business processing system, a vehicle and a storage medium. The method pre-configures different task units and directly calls the target model in the pre-configured model according to the target task unit corresponding to the business consultation request; through unified configuration and management of the model, different business consultation requests can call the model to generate reply content corresponding to the business consultation request; thereby improving the business processing efficiency in the business consultation scenario.

[0078] Figure 1 is an application scenario diagram of the business interaction method provided in the embodiment of the present application; Figure 1 The illustrated application scenario 100 includes a vehicle communicating with a cloud server (i.e., the vehicle manufacturer's cloud server). The cloud server is the vehicle's business support platform, providing vehicle-side services. The cloud server is equipped with a business processing system (also known as the vehicle manufacturer's business system).

[0079] Figure 2 This is a schematic flowchart of a business processing method provided in an embodiment of the present application.

[0080] For example, Figure 2 The method 200 shown can be executed by the car company server (the car company's cloud server) or by the car company's business system; wherein the car company business system is deployed and run in the car company server (that is, the car company server is equipped with the car company business system), and the car company business system represents a business system or technical platform built by the car company based on artificial intelligence big model technology, which has multi-dimensional intelligent processing capabilities. The car company business system is used to handle vehicle-related business. Figure 2 As shown, the business processing method 200 includes S210 to S240, and S210 to S240 are described in detail below.

[0081] S210: Obtain the user's business consultation request.

[0082] Exemplarily, a user's business consultation request is obtained; wherein the user may be a business developer of a car company (for example, an in-vehicle system developer, a smart cockpit designer), or a vehicle user (for example, a driver, a vehicle passenger, etc.). The user's business consultation request may be a business consultation request input in text form. For example, the user inputs "There is an abnormal noise in the vehicle engine, how should I deal with it" in the vehicle computer application. The input content is the user's business consultation request. The business consultation request may also be a business consultation request input in voice form. For example, the user inputs "Locate the nearest charging station" into the voice module of the smart cockpit. The input content is the user's business consultation request. This application does not specifically limit the input form of the business consultation request.

[0083] Exemplary business scenarios for automotive companies include, but are not limited to, smart cockpit interaction, autonomous driving assistance, and connected car services. Specifically, smart cockpit interaction refers to the use of natural language processing to enable voice assistants, scenario-based dialogue, and understanding of user intent, thereby accurately executing vehicle control commands and providing personalized services to users. Autonomous driving assistance combines computer vision, reinforcement learning, and other technologies to enhance perception and planning capabilities in complex road conditions. Connected car services refer to services such as in-car entertainment, intelligent navigation, intelligent question-and-answer, remote vehicle control, and after-sales forecasting based on user profiles and scenario analysis.

[0084] S220: Determine at least two target task units corresponding to the business consulting request.

[0085] Among them, the target task unit is used to execute the task corresponding to the business consultation request, and at least two target task units include an intent recognition task unit and a reply generation task unit. The intent recognition task unit is used to recognize the intent of the user's business consultation request, and the reply generation task unit is used to generate the reply content. It can be understood that the target task unit corresponding to any business consultation request includes an intent recognition task unit and a reply generation task unit. The intent recognition task unit corresponds to the starting point of the business processing flow (the business processing flow corresponding to the business consultation request), and the reply generation task unit corresponds to the end point of the business processing flow, ensuring that the business consultation request input by the user can be parsed and the intent can be recognized, and after processing, the reply content is generated.

[0086] In a specific implementation, the tasks that can be executed by the preconfigured model are split to obtain multiple task units associated with the preconfigured model, wherein different task units are used to execute different tasks; the business consulting request is parsed to determine the target task corresponding to the business consulting request; and the target task unit corresponding to the target task is determined from the multiple task units. Exemplarily, the tasks that can be executed by the preconfigured model are split to obtain multiple task units; that is, the different processing capabilities of different models in the preconfigured model are split and encapsulated to obtain multiple reusable task units (or agents); wherein encapsulation refers to hiding the complex implementation details within the system and exposing only a unified interface that complies with industry or protocol standards to the outside world, which is convenient for external calls, thereby improving modularity and enhancing maintainability.

[0087] Optionally, when splitting and encapsulating task units, a standard definition is provided for each intelligent unit. The definition of an agent includes data such as input parameters, output parameters, the agent call model, and the agent's prompt template and permission control. For example, the agents associated with the large model include a "car model recommendation agent." This agent is defined to receive basic user information and purchase intentions, call the large model to generate recommendation text, and return structured recommendations with confidence levels. By encapsulating the large model, each agent can be reused, scheduled, and managed as a separate module.

[0088] Exemplarily, the user's business consulting request is parsed to obtain the target task corresponding to the business consulting request, wherein parsing the business consulting request refers to the process of parsing the business consulting request input by the user, determining the intention corresponding to the business consulting request and breaking down the intention into multiple step-by-step execution actions, each execution action corresponding to a target task of the business consulting request, that is, the target task of the business consulting request is used to execute the step-by-step execution actions of the business consulting request.

[0089] In one implementation, the process of parsing a user's business consultation request includes: using an intent recognition unit to recognize the intent of the business consultation request based on a preset dictionary and grammatical rules, and breaking down the matched and recognized intent into multiple executable actions.

[0090] For example, user intentions are matched based on preset dictionaries and grammatical rules. For example, "navigate to address A" is identified as "route planning intention"; "mobile phone battery is too low" is identified as "in-vehicle charging port query intention" or "turn on in-vehicle wireless charging intention."

[0091] For example, the matched intent is broken down into multiple executable actions. For example, a user's business inquiry request is "My vehicle has fault A. How can I get it repaired?" The input content is converted into the intent of "Maintenance process planning." This intent is then broken down into multiple target tasks: scheduling a repair, detecting vehicle fault codes, querying nearby repair points, selecting a time, confirming the appointment, and generating a response. By executing these multiple tasks in sequence, the output content corresponding to the user input content is obtained.

[0092] In an embodiment of the present application, the tasks that can be executed by the preconfigured model are split to obtain multiple task units; and the target task unit is determined from the multiple task units according to the target task corresponding to the business consulting request; since each model in the preconfigured model has different processing capabilities and can execute different tasks; therefore, the tasks that can be executed by the preconfigured model are split to obtain task units that can execute different tasks, which facilitates calling the different processing capabilities of the model through the task units; the business consulting request is parsed and decomposed into multiple target tasks, so that the target task unit can be determined from the multiple task units according to the target task, ensuring that the model can be called through the target task unit to execute the target task.

[0093] In one possible implementation, determining whether the user has business consultation authority; and determining at least two target task units corresponding to the business consultation request include: if the user has business consultation authority, determining at least two target task units corresponding to the business consultation request.

[0094] For example, to determine whether a user has business consultation permissions, the user is classified and the permission levels are divided. If the user is a business person in an automobile company (for example, salesperson, after-sales staff, business development personnel), different permissions are assigned according to their functions, and different functions have different business consultation permissions. If the user is a vehicle user, their permissions are generally set to business consultation permissions related to the vehicle itself (for example, querying the vehicle maintenance status, querying the vehicle's fault status). By binding the user identity with the permissions, when it is detected that the user has input a business consultation request, the user's permissions are determined based on the user's identity information carried in the business consultation request, and the business consultation permissions required for the business consultation request are determined, thereby determining whether the user's permissions include business consultation permissions.

[0095] It's important to note that permission rules are typically stored in a permission database or the configuration center of an automaker's business system, taking the form of a "role-resource-operation" triple. Roles represent identity information, resources represent the models and knowledge fragments that can be invoked by that identity, and operations represent the tasks that can be performed by the model that that identity can invoke. For example, a permission rule in the stored triple is "Developer-Recommended Model-Call Query Interface," which indicates that the developer role can invoke the query interface for the recommended model.

[0096] In an embodiment of the present application, since at least two target tasks corresponding to the business consulting request are determined when the user has business consulting authority, it is ensured that each call behavior to the model is controllable and traceable; it prevents users who do not have business consulting authority from calling the model, thereby improving the security of the business consulting service.

[0097] S230, calling a target model in the pre-configured model through the at least two target task units to obtain a target reply content of the business consultation request.

[0098] The pre-configured model includes a plurality of pre-trained large models, such as a deepseek model, a general-purpose language model (GLM), and a model fine-tuned by a vehicle enterprise.

[0099] In an implementation manner, a prompt word template of the at least two target task units is determined; a prompt word is generated based on the business consultation request and the prompt word template; and the prompt word is input into the target model, so that the target model generates the target reply content according to the prompt word.

[0100] For example, the prompt word template of the target task unit is a prompt word template defined and stored when the task unit is split and encapsulated. The user input business consultation request is automatically converted into a standard prompt word through the prompt word template, and then input into the target model, so that the user does not need to manually adapt the template, and the system automatically fills in the variable according to the task to generate a high-quality prompt word, ensuring stable model output effect.

[0101] It can be understood that, since the prompt word is the core medium for user interaction with the model, it has a decisive guiding effect on the content, direction and style of the model output, and directly affects the generation quality when the model generates content. Therefore, the prompt word is generated according to the prompt word template of the target task unit and the business consultation request, so as to generate a high-quality prompt word for the business consultation request, thereby improving the output effect of the target model, i.e., improving the generation quality of the target reply content.

[0102] For example, when the business consultation request is a vehicle fault consultation request, the corresponding target task unit includes task unit A, and a prompt word template of the task unit A is set as "user describes vehicle [fault phenomenon], please analyze possible reasons and provide solutions in combination with common problems of [vehicle model]. If it involves safety risks, the user needs to be prompted to stop using and contact professional maintenance in priority." If it is detected that the user inputs "what should I do when the brake makes a loud noise?" through the vehicle-mounted voice or the customer service system, the user input is converted into a standard prompt word "user describes that the brake of vehicle model A makes a loud noise, please analyze possible reasons and provide solutions. If it involves safety risks, the user needs to be prompted to stop using and contact professional maintenance in priority." according to the prompt word template and the business consultation request of the user.

[0103] It should be noted that the above are examples of target task units and prompt word templates of target task units; this application does not limit the specific task units and prompt word templates.

[0104] In one implementation, the above method also includes: based on the business consultation request, determining whether the business consultation request is associated with a knowledge fragment in the target knowledge base; if the business consultation request is associated with a knowledge fragment in the target knowledge base, inputting the knowledge fragment and prompt words associated with the business consultation request into the target model, so that the target model generates target response content based on the prompt words and the knowledge fragment.

[0105] It is understandable that since existing technical models rely on the knowledge learned during model training when generating content, if relevant knowledge in the business scenario is not injected into the model, the content generated by the model may be inaccurate; therefore, business knowledge related to the business scenario is stored in the target knowledge base; when a business consultation request is associated with a knowledge fragment in the target knowledge base, the knowledge fragments and prompt words in the target knowledge base are input into the target model to ensure that the model can generate more accurate target response content based on the knowledge fragments and prompt words in the target knowledge base.

[0106] For example, a user inquires about the maintenance cycle and maintenance items for a certain car model. Since the maintenance cycles and maintenance items may differ for different car models, in order to ensure the accuracy of the target response content, relevant information is obtained from the target knowledge base corresponding to the business consultation request (the target knowledge base contains knowledge fragments from the company's vehicle maintenance manual). For example, "The first maintenance for this car model is performed after 5,000 kilometers or 6 months of driving, and routine maintenance is performed every 10,000 kilometers or 12 months. Routine maintenance items include changing the engine oil, oil filter, air filter, checking the brake system, tire pressure, etc." These knowledge fragments and prompt words are output to the target model, and the target model generates response content based on the above knowledge fragments and prompt words.

[0107] In one implementation, a business process corresponding to a business consultation request is determined; based on the business process, at least two target task units are sorted to obtain a target task chain; and target models are called in sequence according to the order of at least two target task units in the target task chain.

[0108] Exemplarily, the target task units are sorted according to the business processes corresponding to the business consulting requests to obtain a target task chain; wherein, different business consulting requests correspond to different business processes, and there is a timing dependency between the task units, that is, the output of the previous task unit of two adjacent task units serves as the input of the next task unit.

[0109] Taking the after-sales service business scenario as an example, the process of determining the target task chain is illustrated. When a user requests a repair request with the engine fault light on their vehicle, the user's business inquiry request is an engine repair request. The corresponding business process for vehicle after-sales service is: user intent recognition, fault diagnosis and analysis, repair plan development, parts inventory verification, and repair shop appointment. Each task in the business process corresponds to a task unit. Based on this business process, the target task chain is determined to be {intent recognition unit; fault diagnosis and analysis unit; repair plan development unit; parts inventory verification unit; repair shop appointment unit; response generation unit}.

[0110] Exemplarily, target models are called sequentially based on the order of target task units in the target task chain. The target models can be multiple models corresponding to the target task units. For example, the intent recognition unit receives a user's natural language description as input. The intent recognition unit calls model A to recognize the user's natural language description, converting it into a standardized fault description. The standardized fault description is then sent to the fault diagnosis and analysis unit as input. The fault diagnosis and analysis unit calls model B based on the standardized fault description to generate a fault diagnosis list. This list is then sent to the maintenance plan formulation unit as input to the maintenance plan activation unit. The maintenance plan formulation unit calls model C based on the fault diagnosis list to generate a maintenance plan. This plan is then sent to the parts inventory verification unit as input to the parts inventory verification unit. The parts inventory verification unit verifies the availability of the required parts based on the maintenance plan and outputs the verification results, which are then sent to the maintenance store reservation unit. The maintenance store reservation unit then makes a reservation with a suitable store based on the verification results. Finally, the response generation unit generates and outputs a response.

[0111] It should be noted that the above is an example of the business scenarios and business processes of automobile business, which is used to describe the process of determining the target task chain and calling the model to generate a response based on the target task chain. In actual applications, the business process and task chain can be set according to business needs; this application does not make specific limitations on this.

[0112] In an embodiment of the present application, if the execution order of the target tasks is wrong, it may result in the inability to generate the target reply content or the quality of the generated target reply content is poor. Therefore, the target model is called in sequence according to the sequence of the target task chain to ensure that the target tasks are executed in sequence, thereby generating the target reply content corresponding to the business consultation request according to the correct business process.

[0113] S240, output the target reply content.

[0114] Exemplarily, the generated target reply content is rendered in a user interface (eg, a customer service chat window or an intelligent assistant interface) so that the user can read the reply content corresponding to the business consultation request.

[0115] Optionally, the method also includes: obtaining usage parameters of the preconfigured model, wherein the usage parameters include the calling frequency, response time, failure rate and consumption of text units of the preconfigured model; and performing configuration management on the preconfigured model based on the usage parameters of the preconfigured model.

[0116] For example, if the change in the call frequency of the pre-configured model is greater than the preset change (for example, 100 times / hour), the number of nodes of the model instance is increased to share the call pressure of the model. If the response time of the configured model is longer than the preset time (for example, 500 milliseconds), the user consultation waiting time is long, and the processing tasks that take a long time are located by monitoring and locating them. For the processing tasks that take a long time, high-performance processor nodes are called first or the processing tasks are processed in segments. Alternatively, if the response time is longer than the preset time, processing tasks with higher timeliness requirements are processed first.

[0117] For example, if the failure rate of calling a preconfigured model is greater than a first threshold (e.g., 5%), a low-power standby model of the model is switched to ensure the availability of the basic services of the model, and an alarm is triggered to notify the operation and maintenance personnel. If the consumption of the text unit of the preconfigured model (e.g., token consumption) is greater than a second threshold, the input text is semantically compressed and redundant information is removed to reduce the consumption of the text unit, and a sleep threshold is set for the model with lower usage frequency (e.g., if it is not called for 1 consecutive hour, the resource is released).

[0118] In the embodiments of the present application, the usage parameters of the preconfigured model are obtained and the preconfigured model is managed according to the usage parameters (for example, more computing power is allocated to models with higher call frequencies). This ensures that during the use of the preconfigured model, the preconfigured model can be managed and optimized according to the usage parameters, realizing dynamic scheduling of model resources, fault self-healing, and cost optimization, and ensuring efficient and stable operation of the model in different business scenarios.

[0119] In the above embodiment, after obtaining the user's business consultation request, the target model in the preset configuration model is called according to at least two target task units corresponding to the business consultation request to obtain the target reply content of the business consultation request. Compared with the prior art in which car companies develop large model services for their own business scenarios, this solution pre-configures different task units and directly calls the target model in the pre-configured model according to the target task unit corresponding to the business consultation request. Through the unified configuration and management of the model, different business consultation requests can call the model to generate the reply content corresponding to the business consultation request. By calling the pre-configured model by the task unit to generate the reply content corresponding to the business consultation request in different business scenarios, the problem of low business processing efficiency caused by the business consultant developing models according to different business scenarios in the prior art is solved, thereby improving the business processing efficiency in the business consultation scenario.

[0120] Figure 3 This is a schematic diagram of the system architecture of a business processing system provided in an embodiment of the present application.

[0121] For example, Figure 3 As shown, the business processing system (i.e., the car enterprise business system) includes the platform operation layer, data enhancement layer, service governance layer, intelligent orchestration layer, and model layer. Different layers have different functions. When a user inputs a business consultation request, the target reply content is generated through the different functions of different layers of the car enterprise business system. Figure 2 The method steps in Figure 3 The functions of each level in the system architecture are described in detail.

[0122] The model layer (model base) is pre-configured with a variety of large models, including automaker-developed models and industry-tuned models. The model layer uses high-performance inference engines such as vLLM (an open-source inference engine optimized for large language models), the TensorRT-LLM inference engine (an inference engine optimized for large language models based on a deep learning framework), or the Text Generation Inference (TGI) engine to implement underlying inference services, offering low-latency, high-throughput, and concurrent processing capabilities.

[0123] The vLLM inference engine manages the key-value cache (KV) by page, avoiding memory fragmentation caused by changes in sequence length in traditional inference and significantly improving cache utilization. It also supports streaming generation, returning results while generating text, reducing user waiting time (for example, displaying responses word for word when users inquire). The TensorRT-LLM inference engine, based on a deep learning inference optimizer, performs layer fusion (for example, combining matrix multiplication with activation function calculations), precision quantization, and parallel computing optimization on large language models, reducing computational workload and memory bandwidth requirements. It supports distributed inference across multiple CPUs to improve inference throughput (for example, batch calculations when processing complex technical document retrieval). The TGI inference engine automatically merges batches based on input requests to improve resource utilization. It supports parallel model loading (for example, loading only the model layers required for the conversation), reducing memory usage, and is suitable for edge devices or lightweight deployments (such as local inference in vehicle central control systems).

[0124] For example, when a user inquires about the "battery life of a certain electric car," the system needs to combine the user's historical inquiries (such as the time of purchase and charging habits) and the battery technical documentation to infer the answer. vLLM caches key information from the user's historical conversations to avoid repeated calculations; the TensorRT-LLM quantization model accelerates the calculation of the battery life prediction formula, which can make the reply generation delay less than 200ms. When a user inquires about "the reason why the vehicle fault light is always on," the system needs to retrieve relevant paragraphs from thousands of pages of maintenance manuals and generate answers. TGI dynamically batches multiple users' fault consultation requests and merges inference tasks. TensorRT-LLM uses multiple GPUs to retrieve document fragments in parallel and generate replies in batches, which can increase the throughput to 200 requests / second.

[0125] The model layer is based on a containerized deployment method, combined with an orchestration and scheduling system to achieve dynamic scaling of the model. Computing resources are automatically applied for and released according to the model call load to optimize cost expenditure. Among them, the containerized deployment method refers to encapsulating large model services and their dependencies (for example, framework libraries, configuration files) into lightweight containers to form independently running application units (i.e., task units). Each application unit has the characteristics of environmental isolation, resource limitations, and fast start and stop. Dynamic scaling of the model refers to automatically adjusting the number of containers according to the model call load. When the call volume surges, new containers are automatically created (expanded) to share the request pressure; when the call volume decreases, idle containers are automatically deleted (reduced) to release resources.

[0126] The intelligent orchestration layer (capability encapsulation) is located above the model layer and is used to abstract and encapsulate the large model capabilities of the bottom layer into standardized and orchestrable intelligent service modules. The intelligent orchestration layer adopts an agent orchestration framework to perform secondary encapsulation on the large model capabilities. Under this framework, the platform combines the original large model inference service with external knowledge bases, application program interface tools, search engines and other resources to construct an "intelligent task unit". Through the design idea of "task unit + process orchestration", the model is no longer a static question and answer interface, but a task unit that can complete multi-step logical reasoning, tool calling and information integration.

[0127] For example, to make the task unit more controllable and adaptable, the business processing system designs a unified prompt word template management mechanism for each task unit. Different task units support binding different prompt word templates, dynamically generating high-quality prompts for different business problems, role settings and target tasks, effectively improving the model output effect. And the intelligent orchestration layer integrates the retrieval augmented generation (RAG) mechanism, during the task execution process, the task unit can extract context information from the knowledge base as input supplement, so that the model has the "knowledge generation" ability.

[0128] It can be understood that when traditional businesses access large models, they can only achieve simple interactions such as "asking questions - answering", lack context memory, tool calling, knowledge support, and are difficult to complete complex task chains. The intelligent orchestration layer solves the problem of model capabilities that cannot be combined and the lack of logical structure in the process by constructing task units; at the same time, through the integration of prompt word templates and RAG mechanisms, it can achieve automatic execution and intelligent decision-making of complex business processes.

[0129] The service governance layer (unified interface and access control) is used to provide standardized and unified model calling interfaces for various business parties, and to control the access, calling monitoring and resource quota management of all model services. The service governance layer is the "entrance" and "firewall" for the external opening of model capabilities, ensuring safe access, stable operation and full-process observability of large model services. The service governance layer provides a standard interface through the construction of a unified service gateway system, and all model services in the model capability layer are exposed to the outside through this service governance layer, shielding the differences between the bottom layer models.

[0130] To ensure the security and stability of the service, the business processing system introduces a permission control system in the service governance layer to divide the permissions of users, organizations, and project levels, ensuring that each call behavior of the model is controllable and traceable. The monitoring and log auditing mechanism is embedded in the service governance layer to monitor and record the time delay, status code, response quality, source identity, and other information of model calls and visualize the information to provide support for subsequent operation analysis, performance optimization, and security audit. The quota management mechanism is introduced in the service governance layer to set the upper limit of the number of calls and the frequency limit of users / departments to ensure fair use of resources and avoid single-point abuse.

[0131] It can be understood that the service governance layer realizes the centralized management and unified service of large model capabilities by building a unified, standard, and secure access portal, significantly reduces the access threshold, and improves the overall system security and operation efficiency. It solves the problems of non-uniform interface format, loose permission management, unmonitorable calls, and information leakage in traditional model deployment

[0132] The data enhancement layer (knowledge injection module) is used to inject structured data and unstructured data within an enterprise into the generation process of a large model to improve its contextual understanding ability and answer accuracy. By building a knowledge base and semantic retrieval mechanism, the large model has the ability to "answer with knowledge". By building an enterprise private knowledge base and semantic retrieval system, combined with a vector database and a high-performance retriever to process unstructured data, the data is converted into semantic information that the model can understand, improving the model's understanding and response accuracy to the business. Structured data refers to data that uses predefined and expected formats, and unstructured data refers to data that lacks definition.

[0133] The business processing system uses document parsing and preprocessing modules to clean, segment, and embed unstructured data (such as company manuals, technical documents, PDFs, and web documents), converting the unstructured data into structured data represented by vectors and storing it in a vector database. Using a search mechanism, the business processing system matches semantic similarity between user business inquiries in the vector database, rapidly retrieving the most relevant knowledge fragments to form a contextual input model, thereby enhancing knowledge. The business processing system supports metadata management, such as structural tags, weight classification, and timeliness annotation, for knowledge fragments, ensuring flexible selection of information sources for different business needs. For example, when a user inquires about "reduced battery life of vehicle model A," the system uses tag matching to accurately retrieve knowledge fragments containing the tags "vehicle model A" and "battery system," filtering out information related to unrelated models. Knowledge fragments are labeled by importance through weight classification, and when the knowledge model retrieves knowledge, the results are sorted by weight, ensuring that high-priority information is output first. For example, when a user inquires about "abnormal brake noise," the system prioritizes "safety tips" (weighted 10 points) over "fault causes" (weighted 7 points). Timeliness marking is used to mark the validity period of knowledge. For example, every time a vehicle maintenance manual is updated, the timeliness of the instruction fragments related to the old version of the maintenance manual is marked as expired.

[0134] Optionally, when new documents are detected, the knowledge base is dynamically updated; through the dynamic update mechanism of the knowledge base, it is ensured that new documents can be stored and synchronized in real time without the need to retrain the model, thereby achieving the synchronous evolution of knowledge and business.

[0135] It is understandable that the data enhancement layer solves problems such as the model being out of touch with the business, lack of context, and low content accuracy by injecting structured enterprise knowledge, thereby improving professionalism.

[0136] The platform operation layer (configuration management and continuous evolution) is used to perform operational tasks such as configuration management, call monitoring, version evolution, and grayscale release of model services to ensure the stable operation of large models in daily use and the ability to flexibly adjust and quickly iterate according to business needs. The platform operation layer has a complete operation support system, including business configuration management, model call analysis, model version management, and model grayscale analysis. It supports the rapid adaptation of general large models to automotive business scenarios through lightweight parameter fine-tuning technology.

[0137] For example, by introducing the platform operation layer to centrally manage model resources, task processes, and business configurations, unified operation control and background management are achieved. The platform supports the registration, switching, release, and rollback of different model versions, realizing the lifecycle management of model capabilities.

[0138] The platform operation layer supports business parties to customize prompt word templates, task chains of task units, and knowledge source bindings for different requirements of multiple scenarios and departments. In terms of model iteration, the platform integrates a fine-tuning support module based on a lightweight parameter optimization technology to quickly adapt general models to individual needs without large-scale training power, thus achieving precise output and meeting industry needs. The platform also builds a complete model calling analysis function to visually display indicators such as calling frequency, response time, and failure rate of model services. In addition, the platform operation layer provides a gray release test mechanism (model gray analysis function) and an AB test mechanism to test and optimize models. The gray release test mechanism refers to pushing a system or function to a portion of users before its release to test it in actual environments, collect user feedback, and timely discover and fix potential problems. The AB test mechanism is used to verify the effectiveness of new functions or function changes by randomly allocating users to the original version (group A) and the experimental version (group B) to compare data from the two groups and determine whether the new function has achieved the expected results. Through testing and optimizing models, the platform realizes the continuous evolution and business alignment of model capabilities and reduces long-term operating costs.

[0139] For example, the processing flow of a user's business consultation request is schematically described in combination with the hierarchical architecture of the above business processing system: the user inputs a query request of "How to handle the failure of Over-The-Air technology (OTA) update?" in the application program of the vehicle enterprise or the customer service background. The application program or the customer service background transmits the question to the business processing system through a unified service interface. The application program interface gateway of the service governance layer receives the request, authenticates the user, confirms the query authority corresponding to the query request of the user, records the calling log, assigns a calling identifier, and forwards the query request to the intelligent arrangement layer. The intelligent arrangement layer matches the corresponding task chain (processing flow) according to the query request, and each task unit in the task chain is executed in turn according to the execution order of the task chain: the question content is read, the corresponding task unit (intelligent customer service answering task unit) is matched, and the processing flow of the task unit is triggered to call the basic model reasoning task; each task unit determines whether external knowledge is needed; since the question involves business-specific knowledge such as "OTA update failure", the intelligent agent calls the RAG module: according to the question semantics, relevant content (such as internal materials like OTA Update Guide and Common Fault Handling Manual) is retrieved in the knowledge base of the vehicle enterprise; and the highest matching knowledge fragment is obtained through the data enhancement layer.

[0140] The task unit packages the question and the retrieved knowledge fragments into a unified format, selects a suitable reasoning model (such as the large model in the existing technology or the enterprise's self-developed model), and sends it to the reasoning engine (vLLM or TensorRT-LLM) to perform reasoning generation. Based on the input, the model generates a solution with detailed steps, for example: "Hello, regarding the failure of the OTA update, it is recommended that you check in sequence: confirm whether the network connection is stable; check whether the vehicle version meets the update requirements; clean up the storage space". The reasoning result is returned to the service governance layer, where indicators such as response time and success are recorded and written to the call log. The final reasoning answer is transmitted back to the application or customer service background. The application receives the final answer and renders it on the user interface (such as the customer service chat window or the display interface of the application). The user sees the system reply and the problem is closed-loop solved.

[0141] Figure 4 This is a schematic flowchart of another business processing method provided in an embodiment of the present application.

[0142] Figure 4 The method 400 shown may be executed by a car manufacturer's server (a car manufacturer's cloud server), or may be executed by a car manufacturer's business system.

[0143] like Figure 4 As shown, the service processing method 400 includes S401 to S411, and S401 to S411 are described in detail below.

[0144] S401, obtaining a user's business consultation request.

[0145] Optionally, the implementation of S401 can refer to Figure 2 The relevant description of S210 is not repeated here.

[0146] S402, determine whether the user has business consultation authority; if so, execute S403; if not, execute S404.

[0147] Exemplarily, it is determined whether the user has business consultation authority; if the user has business consultation authority, at least two target task units corresponding to the business consultation request are determined; if the user does not have business consultation authority, a prompt message is output.

[0148] S403: Determine at least two target task units corresponding to the business consultation request.

[0149] For example, tasks that can be executed by the preconfigured model are split into multiple task units. Specifically, the different processing capabilities of different models within the preconfigured model are split and encapsulated to create multiple reusable task units. If the user has business consultation permission, the business consultation request is parsed to determine the target task corresponding to the business consultation request. From the multiple task units, the target task unit corresponding to the target task is determined.

[0150] S404: Output prompt information.

[0151] For example, if the user does not have the business consultation authority, a prompt message is output, where the prompt message is used to remind the user that he does not have the business consultation access authority.

[0152] Optionally, the implementation of S402 to S404 can refer to Figure 2 The relevant description of S220 is omitted here.

[0153] S405 , sorting the target task units according to the business process corresponding to the business consultation request to obtain a target task chain.

[0154] Exemplarily, determine the business process corresponding to the business consultation request; based on the business process, sort at least two target task units (at least two target task units include an intent recognition task unit and a response generation task unit) to obtain a target task chain, wherein the business process is a business processing process set in advance according to business needs.

[0155] S406: Generate prompt words in sequence according to the prompt word templates of the target task chain and the target task unit.

[0156] Exemplarily, a prompt word template of a target task unit is obtained; and according to the business consultation request and the prompt word template, the business consultation request input by the user is automatically converted into a standard prompt word through the prompt word template.

[0157] S407 , determining whether the business consultation request is associated with a knowledge fragment in the target knowledge base; if so, executing S408 ; if not, executing S409 .

[0158] For example, a determination is made as to whether the business consultation request is associated with a knowledge fragment in the target knowledge base. If so, the prompt word and the knowledge fragment are sequentially input into the target model. If not, the prompt word is sequentially input into the target model. The order in which the prompt words are sequentially input into the target model corresponds to the order in which the models are called as shown in the target task chain.

[0159] S408: Input the prompt words and knowledge fragments into the target model in sequence.

[0160] For example, when a business consultation request is associated with a knowledge fragment in the target knowledge base, the knowledge fragment and prompt words in the target knowledge base are input into the target model to ensure that the model can generate more accurate target response content based on the knowledge fragment and prompt words in the target knowledge base.

[0161] S409: Input the prompt words into the target model in sequence.

[0162] Exemplarily, the prompt words are input into the target model in sequence to ensure that the target model can generate responses based on relatively standard prompt words, thereby improving the generation quality of the target model.

[0163] S410, the large model generates target reply content based on the input content.

[0164] For example, if the business consultation request is associated with a knowledge fragment in the target knowledge base, the big model generates the target response content for the business consultation request based on the input prompt words and knowledge fragments; if the business consultation request is not associated with a knowledge fragment in the target knowledge base, the big model generates the target response content corresponding to the business consultation request based on the input prompt words.

[0165] Optionally, the implementation of S405 to S410 can be found in Figure 2 The relevant description of S230 is not repeated here.

[0166] S411, output the target reply content.

[0167] Exemplarily, the generated target reply content is rendered in the user interface so that the user can read the reply content corresponding to the business consultation request.

[0168] Optionally, the implementation of S411 can refer to Figure 2 The relevant description of S240 is not repeated here.

[0169] Combined with the above Figures 1 to 4 The service processing method provided by the embodiment of the present application is described in detail; Figure 5 and Figure 6 The service processing device embodiment of the present application is described in detail. It should be understood that the service processing device in the embodiment of the present application can execute the various methods of the aforementioned embodiments of the present application, that is, the specific working processes of the following various products can refer to the corresponding processes in the aforementioned method embodiments.

[0170] Figure 5 It is a structural diagram of a business processing device provided in an embodiment of the present application.

[0171] For example, Figure 5 As shown, the service processing device 500 includes:

[0172] An acquisition module 510 is used to acquire a user's business consultation request;

[0173] A processing module 520 is configured to determine at least two target task units corresponding to the business consultation request;

[0174] The target model in the preconfigured model is called through at least two target task units to obtain target response content for the business consultation request; and the target response content is output.

[0175] Optionally, as an embodiment, the processing module 520 is specifically used to: determine prompt word templates for at least two target task units; generate prompt words based on the business consultation request and the prompt word template; input the prompt words into the target model so that the target model generates target reply content based on the prompt words.

[0176] Optionally, as an embodiment, the processing module 520 is specifically used to: based on the business consultation request, determine whether the business consultation request is associated with a knowledge fragment in the target knowledge base; if the business consultation request is associated with a knowledge fragment in the target knowledge base, input the knowledge fragment and prompt words associated with the business consultation request into the target model, so that the target model generates target reply content based on the prompt words and the knowledge fragment.

[0177] Optionally, as an embodiment, the processing module 520 is specifically used to: split the tasks that can be executed by the preconfigured model to obtain multiple task units associated with the preconfigured model, wherein different task units are used to execute different tasks; parse the business consulting request to determine the target task corresponding to the business consulting request; and determine the target task unit corresponding to the target task from multiple task units.

[0178] Optionally, as an embodiment, the processing module 520 is also used to: determine the business process corresponding to the business consultation request; based on the business process, sort at least two target task units to obtain a target task chain; and call the target model in sequence according to the order of at least two target task units in the target task chain.

[0179] Optionally, as an embodiment, the processing module 520 is further configured to: determine whether the user has business consultation authority; if the user has business consultation authority, determine at least two target task units corresponding to the business consultation request.

[0180] Optionally, as an embodiment, the acquisition module 510 is also used to: obtain the usage parameters of the preconfigured model, wherein the usage parameters include the calling frequency, response time, failure rate and consumption of text units of the preconfigured model; the processing module 520 is also used to: perform configuration management on the preconfigured model based on the usage parameters of the preconfigured model.

[0181] It should be noted that the above-mentioned service processing device is embodied in the form of a functional unit. The term "module" here can be implemented in the form of software and / or hardware, and is not specifically limited to this.

[0182] For example, a "module" may be a software program, a hardware circuit, or a combination of the two that implements the aforementioned functionality. The hardware circuit may include an application specific integrated circuit (ASIC), an electronic circuit, a processor (e.g., a shared processor, a dedicated processor, or a group of processors) and memory for executing one or more software or firmware programs, combined logic circuits, and / or other suitable components that support the described functionality.

[0183] Therefore, the units of each example described in the embodiments of this application can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians 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.

[0184] Figure 6 It is a structural schematic diagram of a vehicle provided in an embodiment of the present application.

[0185] Exemplarily, vehicle 600 includes a processor 610 , a memory 620 , and executable program code 630 .

[0186] Exemplarily, the vehicle 600 includes one or more processors 610, which can support the vehicle 600 in implementing the service processing method in the method embodiment. The processor 610 can be a general-purpose processor or a special-purpose processor. For example, the processor 610 can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, such as discrete gates, transistor logic devices, or discrete hardware components.

[0187] For example, the processor 610 can be used to control the vehicle 600, execute software programs, and process data of the software programs. The vehicle 600 can also include a communication unit to implement signal input (reception) and output (transmission).

[0188] Exemplarily, the vehicle 600 may include one or more memories 620 on which executable program code 630 is stored. The executable program code 630 can be executed by the processor 610 to generate instructions so that the processor 610 executes the business processing method described in the above method embodiment according to the instructions.

[0189] Optionally, data may be stored in the memory 620. Optionally, the processor 610 may read data stored in the memory 620. The data may be stored at the same storage address as the executable program code 630, or may be stored at a different storage address from the executable program code 630.

[0190] Exemplarily, the processor 610 and the memory 620 may be provided separately or integrated together, for example, integrated on a system on chip (SOC) of the terminal device.

[0191] Exemplarily, the memory 620 can be used to store relevant programs of the business processing method provided in the embodiment of the present application, and the processor 620 can be used to call the executable program code 630 stored in the memory 620 when controlling the vehicle to execute the business processing method of the embodiment of the present application; for example, obtain the user's business consultation request; determine at least two target task units corresponding to the business consultation request; call the target model in the preconfigured model through at least two of the target task units to obtain the target reply content of the business consultation request; and output the target reply content.

[0192] The present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the business processing method of any of the aforementioned embodiments.

[0193] Among them, computer-readable storage media may include, but are not limited to, any type of disk, including floppy disks, optical disks, Digital Video Discs (DVDs), Compact Disc Read-Only Memory (CD-ROMs), microdrives and magneto-optical disks, Read-Only Memory (ROMs), Random Access Memory (RAMs), Erasable Programmable Read-Only Memory (EPROMs), Electrically Erasable Programmable Read-Only Memory (EEPROMs), Dynamic Random Access Memory (DRAMs), Video Random Access Memory (VRAMs), flash memory devices, magnetic or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0194] The present application also provides a computer program product. When the computer program product is run on a computer, it enables the computer to execute the above-mentioned related steps to implement a business processing method in the above-mentioned embodiment.

[0195] In addition, the vehicle provided in the embodiments of the present application can specifically be a chip, component or module, and the vehicle may include a connected processor and memory; wherein the memory is used to store instructions, and when the vehicle is running, the processor can call and execute instructions to enable the chip to execute a business processing method in the above embodiment.

[0196] Among them, the vehicle, computer-readable storage medium, computer program product or chip provided in this application are all used to execute the corresponding business processing methods provided above. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding business processing methods provided above, and will not be repeated here.

[0197] Through the description of the above implementation methods, technical personnel in the relevant field can understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional modules as needed, that is, the internal structure of the business processing device can be divided into different functional modules to complete all or part of the functions described above.

[0198] In the embodiments provided in this application, it should be understood that the disclosed business processing devices and methods can be implemented in other ways. For example, the business processing device embodiments described above are merely illustrative. For example, the division of modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another business processing device, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the business processing devices or units can be electrical, mechanical or other forms.

[0199] The above content is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A business processing method, characterized in that: The processing method comprises: Obtaining users' business consultation requests; Determining at least two target task units corresponding to the business consulting request; Calling a target model in a preconfigured model through at least two of the target task units to obtain target reply content for the business consultation request; Output the target reply content.

2. The method according to claim 1, characterized in that The step of calling a target model in a preconfigured model by at least two target task units to obtain target response content for the business consultation request includes: Determining prompt word templates for at least two of the target task units; generating prompt words based on the business consultation request and the prompt word template; The prompt word is input into the target model so that the target model generates the target response content according to the prompt word.

3. The method according to claim 2, characterized in that The method further comprises: Based on the business consultation request, determining whether the business consultation request is associated with a knowledge fragment in a target knowledge base; Inputting the prompt word into the target model so that the target model generates the target response content according to the prompt word includes: If the business consultation request is associated with a knowledge fragment in the target knowledge base, the knowledge fragment associated with the business consultation request and the prompt word are input into the target model so that the target model generates the target reply content based on the prompt word and the knowledge fragment.

4. The method according to claim 1, wherein The method further comprises: Splitting the tasks that can be executed by the preconfigured model to obtain a plurality of task units associated with the preconfigured model, wherein different task units are used to execute different tasks; The determining of at least two target task units corresponding to the business consultation request includes: Analyze the business consultation request and determine the target task corresponding to the business consultation request; The target task unit corresponding to the target task is determined from the multiple task units.

5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: Determine the business process corresponding to the business consultation request; Based on the business process, sorting the at least two target task units to obtain a target task chain; The calling of the target model in the preconfigured model by at least two of the target task units includes: The target models are called in sequence according to the order of the at least two target task units in the target task chain.

6. The method according to any one of claims 1 to 4, characterized in that The method further comprises: Determining whether the user has business consultation authority; The determining of at least two target task units corresponding to the business consultation request includes: If the user has the business consultation authority, the at least two target task units corresponding to the business consultation request are determined.

7. The method according to any one of claims 1 to 4, characterized in that Also includes: Obtaining usage parameters of the preconfigured model, wherein the usage parameters include call frequency, response time, failure rate, and text unit consumption of the preconfigured model; The preconfigured model is configured and managed based on the usage parameters of the preconfigured model.

8. A business processing system, characterized in that: The business processing system includes an application layer, an intelligent orchestration layer and a model layer; The application layer is used to obtain the user's business consultation request; The intelligent orchestration layer is used to determine at least two target task units corresponding to the business consultation request, and call the target model in the pre-configured model of the model layer through the at least two target task units, so that the target model generates target reply content; The model layer is used to generate the target reply content and send it to the application layer so that the application layer outputs the target reply content.

9. A vehicle, characterized in that: The vehicle comprises: a memory for storing executable program code; A processor is configured to call and run the executable program code from the memory, so that the vehicle implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed, the method according to any one of claims 1 to 7 is implemented.

Images