Service reply method and device, training method and device, equipment and storage medium

Abstract

The invention relates to the technical field of intelligent dialogues, and provides a service reply method, a training method, devices, equipment and a storage medium, and the method comprises the steps: obtaining a preset service problem; based on an information identification sub-model of the service reply model, according to a preset service problem, determining service thinking information and tool selection information corresponding to the preset service problem; the service thinking information comprises service key information; obtaining a tool calling result corresponding to the business key information based on an information obtaining tool corresponding to the tool selection information; and based on a reply sub-model of the service reply model, according to the service thinking information, the tool selection information and the tool calling result, determining reply information for the preset service problem, so as to improve the reply accuracy for the service problem.

Description

Technical Field

[0001] The present application relates to the field of intelligent dialogue technology, and in particular to a business response method, training method, device, equipment and storage medium. Background Art

[0002] At present, since large language models such as the ChatGLM2 model, etc., have powerful natural language processing and generation capabilities, large language models can be used as business response models to respond to corresponding business questions. For example, for the auto insurance query business in the financial field, the business response model can respond to questions related to the auto insurance business raised by users.

[0003] However, in the related art, since the large language model cannot call external tools or query external databases to query user information, and the questions asked by users are generally related to their user information, it is easy to cause poor accuracy in responding to the corresponding business. For example, for the auto insurance inquiry business in the financial field, if the business response model does not accurately respond to questions related to the auto insurance business, it will take extra manpower and time to respond to the user's questions. Summary of the invention

[0004] The main purpose of this application is to provide a business response method, training method, device, equipment and storage medium, aiming to improve the accuracy of responses to business questions.

[0005] In a first aspect, the present application provides a service recovery method, the service recovery method comprising the following steps:

[0006] Get preset business questions;

[0007] Based on the information identification sub-model of the business response model, according to the preset business problem, determine the business thinking information and tool selection information corresponding to the preset business problem; the business thinking information includes business key information;

[0008] Based on the information acquisition tool corresponding to the tool selection information, acquire the tool call result corresponding to the business key information;

[0009] Based on the response sub-model of the business response model, response information to the preset business question is determined according to the business thinking information, the tool selection information and the tool calling result.

[0010] In a second aspect, the present application further provides a training method for a service response model, the training method comprising:

[0011] Obtaining a training sample set, the training sample set including a plurality of preset business problems, business thinking labels, tool selection labels, and tool call result labels corresponding to each of the preset business problems;

[0012] Based on the information identification sub-model of the business response model, according to the preset business problem, determine the business thinking information and tool selection information corresponding to the preset business problem, wherein the business thinking information includes business key information;

[0013] Based on the information acquisition tool corresponding to the tool selection information, acquire the tool call result corresponding to the business key information;

[0014] Determining model optimization parameters of the business response model according to the business thinking information and the business thinking label, the tool selection information and the tool selection label, and the tool call result and the tool call result label;

[0015] According to the model optimization parameters, the model parameters of the business recovery model are adjusted.

[0016] In a third aspect, the present application further provides a service recovery device, the service recovery device comprising:

[0017] Business problem acquisition module, used to obtain preset business problems;

[0018] An information determination module, configured to determine business thinking information and tool selection information corresponding to the preset business problem based on the information identification sub-model of the business response model and according to the preset business problem; the business thinking information includes business key information;

[0019] A tool calling module, used to obtain a tool based on the information corresponding to the tool selection information, and obtain a tool calling result corresponding to the business key information;

[0020] The response generation module is used to determine the response information to the preset business question based on the response sub-model of the business response model according to the business thinking information, the tool selection information and the tool calling result.

[0021] In a fourth aspect, the present application further provides a computer device, the computer device comprising a memory and a processor;

[0022] The memory is used to store computer programs;

[0023] The processor is used to execute the computer program and implement the steps of the business response method as described above, or implement the steps of the business response model training method as described above when executing the computer program.

[0024] In a fifth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the business response method as described above, or implements the steps of the training method of the business response model as described above.

[0025] The present application provides a business response method, training method, device, equipment and storage medium. The business response method includes: obtaining a preset business question; based on the information identification sub-model of the business answer model, determining the business thinking information and tool selection information corresponding to the preset business question according to the preset business question; the business thinking information includes business key information; based on the information acquisition tool corresponding to the tool selection information, obtaining the tool call result corresponding to the business key information; based on the response sub-model of the business response model, determining the response information to the preset business question according to the business thinking information, the tool selection information and the tool call result, so as to improve the accuracy of the response to the business question. For example, it can be used for the auto insurance business in the financial field. Questions related to the auto insurance business can be queried through the business response method. By calling the information acquisition tool corresponding to the tool selection information, the tool call result corresponding to the business key information is obtained, which is conducive to improving the accuracy of the response to the business question. In addition, there is no need to spend a lot of manpower and time to manually respond to questions related to the auto insurance business, which is conducive to improving the operating efficiency of the business question query system. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 It is a flowchart of a service recovery method provided in an embodiment of the present application;

[0028] Figure 2 It is a flowchart of a training method for a business response model provided in an embodiment of the present application;

[0029] Figure 3 is a schematic block diagram of a service recovery device provided in an embodiment of the present application;

[0030] Figure 4 It is a schematic block diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0032] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0033] The embodiments of the present application provide a business response method, training method, apparatus, device and storage medium. The business response method and the training method of the business response model can be applied to a computer device, which can be a mobile phone, a tablet computer, a laptop computer, a desktop computer and other devices. It can also be applied to a server, which can be a separate server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0034] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0035] See also Figure 1 , Figure 1 It is a flowchart of a business response method provided by an embodiment of the present application. It should be noted that the business response method provided by the embodiment of the present application can be used in a computer device, and of course can also be used in a server. For example, the server can obtain a preset business question from a computer device, process the preset business question according to the business response method, and obtain response information for the preset business question. Exemplarily, the server can, for example, send the response information to the computer device; of course, it is not limited to this and is not limited here.

[0036] In specific implementation, computer devices include but are not limited to: any one of mobile phones, tablet computers, laptop computers, and desktop computers; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.

[0037] like Figure 1 As shown, the service recovery method includes steps S101 to S104.

[0038] Step S101: Obtain preset business questions.

[0039] For example, when the business response model detects a corresponding business question input by a user, it can obtain the corresponding business question to determine a preset business question based on the business question. The preset business question can be used to subsequently determine response information to the preset business question.

[0040] Take the vehicle violation information query service, for example, if the user needs to query the vehicle violation information. The user can enter business questions related to querying vehicle violation information in the business question query system. The business response model can, for example, be connected to the business problem query system in the form of a service, and the business response model can perform corresponding business problem detection in the business problem query system. When the business response model detects the business problem related to querying vehicle violation information input by the user, the preset business problem can be determined based on the business problem related to querying vehicle violation information. Among them, the business problem query system may include web pages, applets, applications, etc., which are not limited here.

[0041] In some implementations, an initial business problem is obtained; based on the data processing sub-model of the business response model, data preprocessing is performed on the initial business problem to obtain a preset business problem.

[0042] For example, when the business response model obtains the corresponding business question input by the user, the business response model can determine the business question as the initial business question. Since the initial business question is raised by the user according to the user's own language logic, in order to enable the business response model to better understand the business question that the user wants to query, the data processing sub-model of the business response model can be used to pre-process the initial business question to obtain the preset business question. By pre-processing the data of the initial business question, the user's own language logic corresponding to the initial business question can be converted into the language logic of the business response model corresponding to the preset business question. The business response model can then determine the response information of the preset business question based on the preset business question.

[0043] In this way, the data processing sub-model based on the business response model performs data preprocessing on the acquired initial business questions to obtain preset business questions, which is conducive to subsequently improving the convenience of the business response model in determining the response information of the preset business questions.

[0044] In some implementations, the user privacy information in the initial business question is converted to obtain a preset business question.

[0045] For example, when a user queries an initial business question, the initial business question may involve user privacy information. For example, when a user queries vehicle violation information, the user is required to provide user privacy information such as the license plate number. For another example, when a user queries car insurance information, the user is required to provide user privacy information such as phone number, ID number, facial information, bank card number, etc. Based on the fact that user privacy information can be used to confirm the user's identity and is associated with the response information of the preset business question, the business response model needs to take confidentiality measures for the user's privacy information, so as to prevent the leakage of user privacy information while enabling the business response model to determine the response information of the preset business question based on the processed user privacy information, thereby improving the convenience of responding to business questions and improving the security of user privacy information.

[0046] In order to prevent the leakage of user privacy information, the data processing sub-model of the business response model can convert the user privacy information in the initial business question to obtain the preset business question when processing the initial business question. For example, the data processing sub-model can convert the user privacy information into a string (token). Accordingly, the business response model can determine the initial business question carrying the string as the preset business question.

[0047] In this way, by converting the user privacy information in the initial business question, the preset business question is obtained, which is conducive to improving the security of the user privacy information. Accordingly, the preset business question obtained by converting the user privacy information can still be combined with the converted user privacy information to confirm the reply information of the subsequent preset business question, which is conducive to improving the convenience and accuracy of the reply to the preset business question.

[0048] Step S102: Based on the information identification sub-model of the business response model, according to the preset business problem, determine the business thinking information and tool selection information corresponding to the preset business problem; the business thinking information includes business key information.

[0049] Exemplarily, the business response model is a model with the ability to call an information acquisition tool. For example, the business response model may include a model with the ability to call an information acquisition tool obtained by training the ChatGLM2 model as a basic model, and is certainly not limited to the ChatGLM2 model. For example, it may also include a model with the ability to call an information acquisition tool obtained by training the ChatGPT model as a basic model, and so on, which is not limited here.

[0050] For example, when the business response model obtains a preset business problem, it can call the information identification sub-model to determine the business thinking information and tool selection information corresponding to the preset business problem.

[0051] Take the vehicle violation information query service as an example. The preset business question obtained by the business response model includes, for example, "I want to check whether my car has violated the law." The information recognition sub-model of the business response model can determine that the preset business question involves "car" and "violation" based on the preset business question, and the information recognition sub-model can determine "car" and "violation" as key business information. When the information recognition sub-model determines the business thinking information corresponding to the preset business question, the information recognition sub-model integrates the key business information involved in the preset business question, such as "car" and "violation", determines that the query of vehicle violation information requires obtaining user privacy information, such as license plate number information, and determines the tool that can be called to query vehicle violation information, such as a violation checking tool. Based on this, the information recognition sub-model can determine that the business thinking information corresponding to the preset business question includes, for example, "the user provides the license plate number information, calls the violation checking tool, and feeds back the query results to the user", and the tool selection information corresponding to the preset business question includes, for example, "call the violation checking tool". Of course, the preset business question, the business thinking information corresponding to the preset business question, and the tool selection information are not limited to this, and are not limited here.

[0052] When the information identification sub-model of the business response model determines the business thinking information and tool selection information corresponding to the preset business problem, the business thinking information and tool selection information can be used by the business response model to subsequently determine the response information to the preset business problem.

[0053] In some embodiments, based on the information identification sub-model, key information extraction is performed on the preset business problem to obtain business key information; based on the business key information and the preset tool prompt information corresponding to the information identification sub-model, the business thinking information and tool selection information are determined; the information acquisition tool corresponding to the tool selection information is selected from the information acquisition tools indicated by the preset tool prompt information.

[0054] For example, the information identification sub-model can perform key information extraction processing on the preset business question to obtain key business information. Accordingly, the information identification sub-model can be provided with corresponding preset tool prompt information. The preset tool prompt information can be used to indicate the information acquisition tool that can be called by the business response model. For example, the business response model can perform information acquisition operations on at least one of the internal application of the business query system, the internal database, the external application that has an associated relationship with the business query system, and the external database by calling the information acquisition tool, without limitation here.

[0055] When the information identification sub-model determines the business-critical information, the information identification sub-model can integrate the business-critical information and the information acquisition tool indicated by the preset tool prompt information to determine the business thinking information corresponding to the preset business problem, and make corresponding tool selections for the preset business problem to determine the corresponding tool selection information.

[0056] Take the query service of vehicle violation information as an example. The preset business question may include "Hello, I want to check whether my car has any violation. My license plate number is: [CARID]". Based on the information recognition sub-model, the preset business question is processed by key information extraction, and the obtained business key information includes, for example, "car", "violation", "license plate number [CARID]". The preset tool prompt information corresponding to the information recognition sub-model includes, for example, "You have the following information acquisition tools to call: violation query tool, knowledge base query tool, ...". The information recognition sub-model can consider whether the user needs to provide corresponding user privacy information based on business key information such as "car", "violation", "license plate number [CARID]", and consider whether the corresponding information acquisition tool needs to be selected from the information acquisition tools indicated by the preset tool prompt information. The information recognition sub-model combines the business key information and the preset tool prompt information, and determines the corresponding business thinking information, for example, including "the user provides the license plate number information, calls the violation query tool, and feeds back the query results to the user.". Accordingly, the tool selection information corresponding to the preset business question determined by the information recognition sub-model includes, for example, selecting the violation query tool from the violation query tool and the knowledge base query tool.

[0057] Of course, this is not limited to this. In the case where the user does not provide the license plate number information, the business response model can output a default response message, such as "Please provide the license plate number information, and I will help you check it from the background." Of course, the default response information is not limited to this, and there is no limitation here. The user can enter the license plate number information in response to the preset response information of the business response model. Accordingly, the information recognition sub-model of the business response model can determine the license plate number information input by the user as the content of the preset business question, and determine the license plate number information as the business key information of the preset business question, so that the information recognition sub-model can combine the license plate number information, which is the business key information, to determine the business thinking information and tool selection information corresponding to the preset business question, and there is no limitation here.

[0058] In an exemplary implementation, a vehicle violation information query service is used as an example. The service response model may be provided with a large model prompt information instruction. The large model prompt information instruction may include preset tool prompt information. When the service response model obtains a preset service question such as "My license plate number is: [CARID]", the service response model may determine the service thinking information and tool selection information corresponding to the preset service question by:

[0059] {

[0060] "instruction": "You are an after-sales agent of an insurance company, and you are good at answering users' questions about policies, claims, revisions, and other auto insurance-related issues. Please answer users' questions in a friendly and patient manner, and do not be rude or disrespectful. You have the following tools to call: violation_query violation query tool; faq knowledge base query tool",

[0061] "input": "<|Human|>: My license plate number is: [CARID]",

[0062] "output": "<|Thought|>: The user provides the license plate number information, calls the violation query tool, and feeds back the query results to the user. <|Commands|>: violation_query(car_id=CARID)",

[0063] "histories": [ [

[0065] "Hello, I want to check if my car has any violations."

[0066] "Hello, it's a pleasure to serve you. Please provide the following license plate number and I will check it for you from the backend." ] ]

[0069] }

[0070] Among them, input is used to indicate the input information of the information identification submodel, such as a preset business problem. Human is used to indicate the user who inputs the preset business problem. Output is used to indicate the output information of the information identification submodel, such as business thinking information and tool selection information. Thought is used to indicate the business thinking information output by the information identification submodel. Commands is used to indicate the tool selection information output by the information identification submodel. CARID is used to indicate the business key information included in the business thinking information. Histories is used to indicate the initial business dialogue associated with the preset business problem. The initial business dialogue may include other preset business problems in addition to the preset business problem.

[0071] Based on the information identification sub-model of the business response model, it can be determined that the business thinking information corresponding to the preset business question includes "the user provides the license plate number information, calls the violation check tool, and feeds back the query result to the user.", and the tool selection information includes calling the violation check tool. Correspondingly, it can also be determined that the business key information includes the license plate number CARID. Of course, it is not limited to the query business of vehicle violation information, and is not limited here.

[0072] In this way, based on the information identification sub-model, the business key information of the preset business problem is determined, and the business key information can be used by the information identification sub-model in combination with the preset tool prompt information to determine the business thinking information and tool selection information, which is conducive to improving the convenience and accuracy of determining the business thinking information and tool selection information corresponding to the preset business problem. The business thinking information and tool selection information corresponding to the preset business problem can be used by the business response model to subsequently determine the response information to the preset business problem, which is conducive to subsequently improving the accuracy of the response to the preset business problem.

[0073] Step S103: Obtain a tool based on the information corresponding to the tool selection information to obtain a tool call result corresponding to the business key information.

[0074] Exemplarily, the tool selection information can be used to instruct the business response model to determine the information acquisition tool required to be called for the response information of the preset business question. For example, when the preset business question indicates that the user needs to query the vehicle violation information, the tool selection information determined by the information identification sub-model of the business response model can be used to instruct the business response model to call the violation query tool. When the preset business question indicates that the user needs to query the knowledge base, the tool selection information determined by the information sub-model of the business response model can be used to instruct the business response model to call the knowledge base query tool. Of course, this is not limited to this and is not limited here.

[0075] In some embodiments, the business response model may call the information acquisition tool corresponding to the tool selection information based on the indication of the tool selection information. In the process of the business response model calling the information acquisition tool corresponding to the tool selection information, the information acquisition tool may obtain the corresponding tool call result based on the business key information. Take the vehicle violation information query business as an example. The information acquisition tool corresponding to the tool selection information may include a violation query tool. The violation query tool may obtain the corresponding vehicle violation information based on business key information, such as license plate number information, and determine the tool call result based on the obtained vehicle violation information. For example, in the case where the vehicle violation information obtained by calling the violation query tool is that no vehicle violation information is obtained, the business response model may determine that the tool call result is no violation. For another example, in the case where the vehicle violation information obtained by calling the violation query tool includes at least one vehicle violation information, the business response model may determine the tool call result based on the at least one vehicle violation information.

[0076] In this way, when the corresponding information acquisition tool can be called based on the tool selection information, it is helpful to improve the accuracy of calling the information acquisition tool. In addition, when the tool call result corresponding to the business key information can be obtained by calling the corresponding information acquisition tool, the tool call result can be used by the business response model to subsequently determine the response information to the preset business question, which is helpful to subsequently improve the accuracy of the response to the preset business question.

[0077] Step S104: Based on the response sub-model of the business response model, the response information to the preset business question is determined according to the business thinking information, the tool selection information and the tool calling result.

[0078] For example, when the business thinking information, tool selection information and tool call results corresponding to the preset business problem are determined, the response sub-model of the business response model can comprehensively consider the preset business problem, business thinking information, tool selection information and tool call results to confirm the response information of the preset business problem.

[0079] In some implementations, preset business questions, business thinking information, tool selection information, and tool call results are input into a response sub-model to obtain response information for the preset business questions.

[0080] For example, when the reply sub-model obtains the preset business problem, business thinking information, tool selection information, and tool call results, the reply sub-model can determine the reply information of the preset business problem according to at least one of the preset business problem, business thinking information, tool selection information, and tool call results. Correspondingly, the reply sub-model can also splice at least two of the preset business problem, business thinking information, tool selection information, and tool call results to obtain the reply information of the preset business problem.

[0081] Take the vehicle violation information query business as an example. Preset business questions include, for example, "I want to check if my car has any violations." and "My license plate number is: [CARID]". Business thinking information includes, for example, "The user provides the license plate number information, calls the violation query tool, and feeds back the query results to the user." Tool selection information includes, for example, calling the violation query tool. The tool call result includes, for example, no violation. The reply sub-model can feed back the tool call result as the reply information of the preset business question to the user according to the logic indicated by the business thinking information. Of course, it is not limited to this. The reply sub-model can also generate reply information such as "The query result of the vehicle violation information of the license plate number [CARID] is no violation" by splicing the business key information included in the business thinking information and the tool call result. The reply sub-model can also generate reply information such as "No violation information has been found through the violation query tool" by splicing the tool selection information and the tool call result, etc., which are not limited here.

[0082] In an exemplary implementation, take the query service of vehicle violation information as an example. The service response model may be provided with a large model prompt information instruction. The large model prompt information instruction may include preset tool prompt information. When the service response model obtains a preset service question such as "Hello, I want to check whether my car has violated the traffic rules.", the service response model may determine the tool call result corresponding to the key service information by:

[0083] {

[0084] "instruction": "You are an after-sales agent of an insurance company, and you are good at answering users' questions about policies, claims, revisions, and other auto insurance-related issues. Please answer users' questions in a friendly and patient manner, and do not be rude or disrespectful. You have the following tools to call: violation_query violation query tool; faq knowledge base query tool",

[0085] "input": "<|Human|>: Hello, I want to check if my car has any violations. <|Thought|>: The user provides the license plate number information, calls the violation check tool, and feeds back the query results to the user. <|Commands|>: violation_query(car_id=CARID). <|Results|>: No violations",

[0086] "output": <|Output|>: Hello, I checked the background and found that your car has no violation information. <eoo>\n",

[0087] "histories": [ [

[0089] "Hello, I want to check if my car has any violations."

[0090] "Hello, it's a pleasure to serve you. Please provide the following license plate number and I will check it for you from the backend." ] ]

[0093] }

[0094] Among them, input is used to indicate the input information of the business response model, such as preset business questions, business thinking information and tool selection information. The business thinking information includes business key information, such as the license plate number CARID, and the tool call result corresponding to the business key information. Output is used to indicate the response information of the business response model to the preset business question. Thought is used to indicate business thinking information. Commands is used to indicate tool selection information. Results is used to indicate the tool call result. CARID is equivalent to the business key information included in the business thinking information, such as the license plate number. Histories is used to indicate the initial business dialogue associated with the preset business question. The initial business dialogue may include other preset business questions in addition to the preset business question.

[0095] By calling the violation checking tool corresponding to the tool selection information, it can be determined that the violation checking tool calling result corresponding to the license plate number CARID is no violation, and the business response model can output the response information to the preset business question based on the tool calling result. Of course, it is not limited to the vehicle violation information query business, and is not limited here.

[0096] In this way, the business response model determines the response information to the preset business question through the response sub-model according to the business thinking information, tool selection information and tool call results. When the user views the response information of the preset business question, the user can view at least one of the business thinking information, tool selection information and tool call results of the business response model for the preset business question, which is conducive to improving the accuracy of the response to the preset business question, and is also conducive to improving the persuasiveness of the response information to the user, thereby helping to improve the user experience of the business problem query system.

[0097] The business response method provided in the above embodiment obtains a preset business question; based on the information identification sub-model of the business answer model, determines the business thinking information and tool selection information corresponding to the preset business question according to the preset business question; the business thinking information includes business key information; based on the information acquisition tool corresponding to the tool selection information, obtains the tool call result corresponding to the business key information; based on the response sub-model of the business response model, determines the response information to the preset business question according to the business thinking information, the tool selection information and the tool call result, so as to improve the accuracy of the response to the business question.

[0098] See also Figure 2 , Figure 2 It is a flow chart of a training method for a business response model provided in an embodiment of the present application. It should be noted that the training method for the business response model provided in an embodiment of the present application can be used in a computer device, and of course can also be used in a server. For example, the server can obtain a training sample set from a computer device, process the training sample set according to the training method of the business response model, and obtain model optimization parameters for the business response model. And according to the model optimization parameters, adjust the model parameters of the business response model to obtain an optimized business response model. Exemplarily, the server can, for example, send the optimized business response model to a computer device, so that when the computer device obtains a preset business problem, it can determine the response information to the preset business problem through the optimized business response model; of course, it is not limited to this, and no limitation is made here.

[0099] In specific implementation, computer devices include but are not limited to: any one of mobile phones, tablet computers, laptop computers, and desktop computers; the server can be a single server or a server cluster, or a cloud server that provides cloud computing services.

[0100] like Figure 2 As shown, the training method of the business response model includes steps S201 to S205.

[0101] Step S201: Obtain a training sample set, where the training sample set includes a plurality of preset business problems, business thinking labels, tool selection labels, and tool call result labels corresponding to each preset business problem.

[0102] For example, when training a business response model, a training sample set may be obtained. The training sample set may include multiple preset business problems. Each preset business problem in the training sample set may be provided with a corresponding business thinking label, a tool selection label, and a tool call result label. The training sample set may be used for subsequent training of the business response model.

[0103] In some implementations, an initial business conversation is obtained; role information corresponding to the initial business conversation is added to the initial business conversation; based on the role information, conversation length information and conversation turn information in the initial business conversation are filtered out to obtain an initial business question; user privacy information in the initial business question is converted to obtain a preset business question.

[0104] Take the vehicle violation information query service as an example. The business query system associated with the vehicle violation information query service can collect a large number of conversations between users and service personnel regarding vehicle violation information. When obtaining multiple preset business questions included in the training sample set, the conversations collected by the business query system can be processed accordingly to determine the preset business questions. For example, the conversations collected by the business query system can be determined as the initial business conversation.

[0105] The obtained initial business dialogue includes, for example, "Hello, I want to check whether my car has any violations. Hello, I am happy to serve you. Please provide the license plate number and I will check it for you from the background. My license plate number is: [car_id]. Hello, I checked from the background and found that your car has no violation information."

[0106] After adding the corresponding role information, the initial business dialogue includes, for example, "User: Hello, I want to check whether my car has any violations. Agent: Hello, I am happy to serve you. Please provide the license plate number and I will check it for you from the background. User: My license plate number is: [car_id]. Agent: Hello, I checked from the background and found that your car has no violation information."

[0107] Accordingly, the conversation lengths of the user and the agent in different conversation rounds can be determined according to the role information. The conversation length information and conversation round information in the initial business conversation can be determined according to the conversation lengths of the user and the agent in different conversation rounds. Based on this, the conversation length information and conversation round information in the initial business conversation can be filtered out according to the role information to obtain the initial business question. The initial business question includes, for example, at least one of "I want to check if my car has violated the law" and "My license plate number is: [car_id]".

[0108] Since the user's license plate number is user privacy information, the user privacy information in the initial business question can also be converted to protect the user privacy information. For example, [car_id] is converted to obtain [CARID]. Accordingly, it can be determined that the preset business question includes at least one of "I want to check whether my car has violated the law" and "My license plate number is: [CARID]". There is an association relationship between the latter preset business question and the former preset business question.

[0109] Of course, the initial business dialogue, initial business questions, and preset business questions are not limited to these and are not restricted here.

[0110] In this way, when the preset business questions are determined, the preset business questions can be used for subsequent training of the business response model.

[0111] In some implementations, when a preset business problem is determined, a business thinking label, a tool selection label, and a tool call result label corresponding to the preset business problem may be determined by a preset marking method.

[0112] Take the query business of vehicle violation information as an example. When it is determined that the preset business question includes at least one of "I want to check whether my car has violated the law" and "My license plate number is: [CARID]", the business thinking label, tool selection label and tool call result label corresponding to the preset business question can be determined.

[0113] For example, for the preset business question "I want to check whether my car has violated any traffic rules", the business thinking tags include, for example, "the user queries the traffic violation scenario, asks the user for the license plate number, and then checks it from the background", the tool selection tags include, for example, "none", and the tool call result tags include, for example, "none". For the preset business question "My license plate number is: [CARID]", since this preset business question is related to the previous preset business question, it can be determined that the business thinking tags include, for example, "the user provides the license plate number information, calls the violation check tool, and feeds back the query results to the user", the tool selection tags include, for example, "traffic violation query tool", and the tool call result tags include, for example, "no violation".

[0114] Of course, the preset business problems, the business thinking tags corresponding to the preset business problems, the tool selection tags, and the tool call result tags are not limited to these and are not restricted here.

[0115] In this way, when the business thinking labels, tool selection labels, and tool call result labels corresponding to the preset business problems are determined, the business response model can be trained subsequently.

[0116] Step S202: Based on the information identification sub-model of the business response model, according to the preset business problem, determine the business thinking information and tool selection information corresponding to the preset business problem, where the business thinking information includes key business information.

[0117] For example, when training the business response model, it is necessary to learn how to determine the business thinking information and tool selection information corresponding to the preset business problem. For example, it is necessary to learn how to identify the business key information included in the preset business problem, and based on the business key information, learn how to generate the corresponding business thinking information and tool selection information. Based on this, the preset business problem can be input into the information recognition sub-model of the business response model, so that the information recognition sub-model can determine the business thinking information and tool selection information corresponding to the preset business problem, and the business thinking information includes the business key information.

[0118] In this way, after determining the business thinking information and tool selection information corresponding to the preset business problem, the business response model can integrate the business thinking information and business thinking labels, as well as the tool selection information and tool selection labels, to optimize the model parameters of the business response model to improve the training effect of the business response model.

[0119] Step S203: Obtain a tool based on the information corresponding to the tool selection information to obtain a tool call result corresponding to the business key information.

[0120] For example, when training the business response model, it is necessary to learn how to obtain tool call results. For example, it is necessary to learn how to call the information acquisition tool corresponding to the tool selection information based on the business key information to obtain the tool call result. Based on this, the tool call result corresponding to the business key information can be obtained based on the information acquisition tool corresponding to the tool selection information. For example, the call result of the violation check tool corresponding to the license plate number is obtained.

[0121] In this way, when the tool call results corresponding to the business key information are determined, the business response model can integrate the tool call results and the tool call result labels to optimize the model parameters of the business response model to improve the training effect of the business response model.

[0122] Step S204: Determine the model optimization parameters of the business response model according to the business thinking information and the business thinking label, the tool selection information and the tool selection label, and the tool call result and the tool call result label.

[0123] For example, the business thinking information, tool selection information, and tool call results determined by the business response model are equivalent to the predicted results of the business response model. The business thinking labels, tool selection labels, and tool call result labels corresponding to the preset business problems are equivalent to the actual results of the business response model. According to the business thinking information and the business thinking labels, a first loss value between the actual result and the predicted result of the business response model when determining the business thinking information can be determined. According to the tool selection information and the tool selection labels, a second loss value between the actual result and the predicted result of the business response model when determining the tool selection information can be determined. Accordingly, according to the tool call results and the tool call result labels, a third loss value between the actual result and the predicted result of the business response model when determining the tool call results can be determined.

[0124] The business response model can determine the model optimization parameters of the business response model according to the first loss value, the second loss value, and the third loss value. The model optimization parameters can be used to subsequently adjust the model parameters of the business response model, so that the business response model can more accurately bring the business thinking information, tool selection information, and tool call results corresponding to the preset business problem of the business response model closer to the actual results, thereby achieving the purpose of training.

[0125] Step S205: Adjust the model parameters of the service response model according to the model optimization parameters.

[0126] For example, the business response model can adjust the model parameters of the information identification sub-model of the business response model according to the model optimization parameters. Of course, it is not limited to this. The business response model can also adjust the model parameters of the response sub-model, data processing sub-model, etc. of the business response model according to the model optimization parameters. This is not limited here.

[0127] In some embodiments, the business response model includes a model designed based on the ChatGLM2 model, and the model parameters of the business response model can be adjusted using LoRA, P-tuning v2, etc. For example, when adjusting the parameters of the business response model, the business thinking information, tool selection information, tool call results, etc. corresponding to the preset business problem can be used to fine-tune the business response model to optimize the performance of the business response model in determining the business thinking information, tool selection information, and tool call results, and to increase the calling capability of the information acquisition tool for the business response model. Of course, it is not limited to this, and no limitation is made here.

[0128] By adjusting the model parameters of the business response model, the business response model can more accurately bring the business thinking information, tool selection information, and tool call results corresponding to the preset business problem closer to the actual results, thereby achieving the training purpose of the business response model. Accordingly, when the business thinking information, tool selection information, and tool call results are closer to the actual results, the business response model can subsequently respond to the preset business problem more accurately, which is conducive to improving the accuracy of the business response model's response to the preset business problem.

[0129] The training method of the business response model provided in the above embodiment obtains a training sample set, the training sample set includes multiple preset business problems, business thinking labels, tool selection labels and tool call result labels corresponding to each preset business problem; based on the information identification sub-model of the business response model, the business thinking information and tool selection information corresponding to the preset business problem are determined according to the preset business problem, and the business thinking information includes business key information; based on the information acquisition tool corresponding to the tool selection information, the tool call result corresponding to the business key information is obtained; according to the business thinking information and the business thinking label, the tool selection information and the tool selection label, and the tool call result and the tool call result label, the model optimization parameters of the business response model are determined; according to the model optimization parameters, the model parameters of the business response model are adjusted to improve the accuracy of determining the business thinking information, tool selection information and tool call results when performing business responses, thereby improving the training effect of the business response model. The business response model can be applied to responding to preset business problems in the future, which is conducive to improving the accuracy of responding to preset business problems. For example, it can be used for auto insurance business in the financial field. Questions related to auto insurance business can be queried through this business response method. By calling the information acquisition tool corresponding to the tool selection information and obtaining the tool call results corresponding to the business key information, it is helpful to improve the accuracy of the response to business questions. In addition, there is no need to spend a lot of manpower and time to manually respond to questions related to the auto insurance business, which is helpful to improve the operating efficiency of the business question query system.

[0130] See also Figure 3 , Figure 3 1 is a schematic block diagram of a service recovery device provided in an embodiment of the present application. The service recovery device can be configured in a server or an electronic device to execute the aforementioned service recovery method.

[0131] like Figure 3 As shown, the business response device includes: a business problem acquisition module 110, an information determination module 120, a tool calling module 130 and a response generation module 140.

[0132] The business problem acquisition module 110 is used to acquire preset business problems.

[0133] The information determination module 120 is used to determine the business thinking information and tool selection information corresponding to the preset business problem based on the information identification sub-model of the business response model and according to the preset business problem; the business thinking information includes business key information.

[0134] The tool calling module 130 is used to obtain a tool based on the information corresponding to the tool selection information, and obtain a tool calling result corresponding to the business key information.

[0135] The response generation module 140 is used to determine the response information to the preset business question based on the response sub-model of the business response model, according to the business thinking information, the tool selection information and the tool calling result.

[0136] It should be noted that technicians in the relevant field can clearly understand that for the convenience and simplicity of description, the specific working processes of the above-described devices and modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0137] The method of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0138] Exemplarily, the above method and apparatus may be implemented in the form of a computer program, which may be run on a computer device.

[0139] See also Figure 4 , Figure 4 1 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server or an electronic device.

[0140] like Figure 4 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and an internal memory.

[0141] The storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any service recovery method, or can execute any service recovery model training method.

[0142] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0143] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute any business recovery method, or can execute any business recovery model training method.

[0144] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0145] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0146] In one embodiment, the processor is used to execute a computer program and implement the following steps when executing the computer program:

[0147] Get preset business questions;

[0148] Based on the information identification sub-model of the business response model, according to the preset business problem, determine the business thinking information and tool selection information corresponding to the preset business problem; the business thinking information includes business key information;

[0149] Based on the information acquisition tool corresponding to the tool selection information, acquire the tool call result corresponding to the business key information;

[0150] Based on the response sub-model of the business response model, response information to the preset business question is determined according to the business thinking information, the tool selection information and the tool calling result.

[0151] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and brevity of description, the specific working process of the business response described above can refer to the corresponding process in the aforementioned business response method embodiment, and will not be repeated here.

[0152] In another embodiment, the processor is used to execute a computer program and implement the following steps when executing the computer program:

[0153] Obtaining a training sample set, the training sample set including a plurality of preset business problems, business thinking labels, tool selection labels, and tool call result labels corresponding to each of the preset business problems;

[0154] Based on the information identification sub-model of the business response model, according to the preset business problem, determine the business thinking information and tool selection information corresponding to the preset business problem, wherein the business thinking information includes business key information;

[0155] Based on the information acquisition tool corresponding to the tool selection information, acquire the tool call result corresponding to the business key information;

[0156] Determining model optimization parameters of the business response model according to the business thinking information and the business thinking label, the tool selection information and the tool selection label, and the tool call result and the tool call result label;

[0157] According to the model optimization parameters, the model parameters of the business recovery model are adjusted.

[0158] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the training of the business response model described above can refer to the corresponding process in the aforementioned business response model training method embodiment, and will not be repeated here.

[0159] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. The method implemented when the computer program is executed by the processor can refer to the various embodiments of the business response method of the present application, or can refer to the various embodiments of the training method of the business response model of the present application.

[0160] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.

[0161] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms of "a", "an" and "the" are intended to include plural forms.

[0162] It should also be understood that the term "and / or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.

[0163] The serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.< / eoo>

Claims

1. A service recovery method, characterized in that: include: Get preset business questions; Based on the information identification sub-model of the business response model, according to the preset business problem, determine the business thinking information and tool selection information corresponding to the preset business problem; the business thinking information includes business key information; Based on the information acquisition tool corresponding to the tool selection information, acquire the tool call result corresponding to the business key information; Based on the response sub-model of the business response model, response information to the preset business question is determined according to the business thinking information, the tool selection information and the tool calling result.

2. The service recovery method according to claim 1, characterized in that: The information identification sub-model based on the business response model determines the business thinking information and tool selection information corresponding to the preset business problem according to the preset business problem, including: Based on the information identification sub-model, key information extraction processing is performed on the preset business problem to obtain key business information; The business thinking information and the tool selection information are determined according to the business key information and the preset tool prompt information corresponding to the information identification sub-model; the information acquisition tool corresponding to the tool selection information is selected from the information acquisition tools indicated by the preset tool prompt information.

3. The service recovery method according to claim 2, characterized in that: The response sub-model based on the business response model determines the response information to the preset business question according to the business thinking information, the tool selection information and the tool calling result, including: The preset business problem, the business thinking information, the tool selection information and the tool calling result are input into the response sub-model to obtain response information of the preset business problem.

4. The method for replying a conversation according to any one of claims 1 to 3, characterized in that: The obtaining of preset business problems includes: Get the initial business problem; Based on the data processing sub-model of the business response model, data preprocessing is performed on the initial business problem to obtain the preset business problem.

5. The service recovery method according to claim 4, characterized in that: The performing data preprocessing on the initial business problem to obtain the preset business problem includes: The user privacy information in the initial business question is converted to obtain the preset business question.

6. A training method for a business response model, characterized in that: The training method comprises: Obtaining a training sample set, the training sample set including a plurality of preset business problems, business thinking labels, tool selection labels, and tool call result labels corresponding to each of the preset business problems; Based on the information identification sub-model of the business response model, according to the preset business problem, determine the business thinking information and tool selection information corresponding to the preset business problem, wherein the business thinking information includes business key information; Based on the information acquisition tool corresponding to the tool selection information, acquire the tool call result corresponding to the business key information; Determining model optimization parameters of the business response model according to the business thinking information and the business thinking label, the tool selection information and the tool selection label, and the tool call result and the tool call result label; According to the model optimization parameters, the model parameters of the business recovery model are adjusted.

7. The training method according to claim 6, characterized in that: The step of obtaining a training sample set includes: Get initial business conversations; Adding role information corresponding to the initial business dialogue to the initial business dialogue; According to the role information, filter out the conversation length information and conversation turn information in the initial business conversation to obtain the initial business question; The user privacy information in the initial business question is converted to obtain a preset business question.

8. A service recovery device, characterized in that: The service recovery device comprises: Business problem acquisition module, used to obtain preset business problems; An information determination module, configured to determine business thinking information and tool selection information corresponding to the preset business problem based on the information identification sub-model of the business response model and according to the preset business problem; the business thinking information includes business key information; A tool calling module, used to obtain a tool based on the information corresponding to the tool selection information, and obtain a tool calling result corresponding to the business key information; The response generation module is used to determine the response information to the preset business question based on the response sub-model of the business response model according to the business thinking information, the tool selection information and the tool calling result.

9. A computer device, characterized in that: The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the dialogue response method as described in any one of claims 1 to 5 when executing the computer program, or implement the dialogue response model training method as described in any one of claims 6 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the dialogue response method as described in any one of claims 1 to 5 are implemented, or the steps of the dialogue response model training method as described in any one of claims 6 to 7 are implemented.

Images