Service order throwing acceptance method and device and medium

Abstract

The invention provides a business rejection acceptance method and device and a medium, and relates to the technical field of artificial intelligence. The method is applied to a service order throwing system or a service order throwing acceptance system comprising a service acceptance system and a service order throwing system, and the method at least comprises the following steps executed by the service order throwing system: obtaining a slot position and slot position configuration of a to-be-handled service; according to the slot position and the slot position configuration, conversation with the user is carried out based on a large language model to complete slot position information filling; and fusing the information filled in all the slots of the to-be-handled service to generate a service acceptance message, and sending the service acceptance message to a service acceptance system. According to the method and the device, the slot position in the business order throwing acceptance process is designed, the big language model is used for dialogue with the user based on the slot position, the user is guided to provide necessary information of business acceptance, the business acceptance message is automatically generated through slot position filling, and intelligent acquisition of the business acceptance information and an efficient and accurate automatic business processing flow are realized.

Description

Technical Field

[0001] This application is at least related to the field of artificial intelligence technology, and particularly relates to a method, device, and medium for accepting business transfer orders. Background Art

[0002] In the existing business acceptance forms, especially for the acceptance forms of government and enterprise services of operators, it mainly relies on business personnel to enter the information on the page line by line according to the business acceptance page and then generate business orders. This method of entering page text has the following deficiencies in use: business personnel need to be familiar with the page entry and page functions, and understand the business meaning represented by each element; using the methods of text and page selection for entry, the entry efficiency depends on the business proficiency of business personnel. This makes business personnel need to learn each new business in order to improve the processing speed of business acceptance, which affects the overall work efficiency and service quality. Summary of the Invention

[0003] In view of the above deficiencies, this application provides a method, device, and medium for accepting business transfer orders to solve the following technical problems: how to simplify the business acceptance process by combining artificial intelligence technology.

[0004] In a first aspect, this application provides a method for accepting business transfer orders. The method is applied to a business transfer order system, or a business transfer order acceptance system including a business acceptance system and a business transfer order system. The method at least includes the following steps executed by the business transfer order system:

[0005] Obtain the slots and slot configurations of the business to be processed;

[0006] Based on the slots and slot configurations, conduct a conversation with the user based on a large language model to complete the filling of slot information;

[0007] Fuse the information filled in all the slots of the business to be processed to generate a business acceptance message, and send the business acceptance message to the business acceptance system.

[0008] Further, before obtaining the slots and slot configurations of the business to be processed, the method further includes the following steps executed by the business transfer order system:

[0009] Pre-design business acceptance message templates corresponding to multiple business scenarios. Each business acceptance message template has business elements to be collected and filled;

[0010] Take each business element to be collected and filled as a slot, and configure the collection and filling rules for each slot information;

[0011] Group the slots according to the business scenarios and collection and filling rules, so that the slots in the same group can complete information collection and filling into the same business acceptance message template at one time.

[0012] Further, group the slots according to the business scenario and the collection and filling rules, so that the slots in the same group can complete information collection at one time and be filled into the same business acceptance message template, specifically including:

[0013] According to the collection and filling rules, the slots are at least divided into three categories: the first slot, the second slot, and the third slot, and each category of slots is divided into several groups according to the involved business acceptance message template;

[0014] Design a basic information guidance template for the business to be handled according to the first slot. The basic information guidance template is used to guide the user to input the basic information of the business to be handled and can at least identify the business scenario of the business to be handled;

[0015] Design prompt words for generating user questions based on the large language model according to each group of second slots, so that the large language model generates a user question for collecting the information of each group of second slots according to the prompt words;

[0016] Design component cards according to each group of third slots. The component cards are used to restrict the user to input the information to be filled into the third slot in a selected manner.

[0017] Further, obtain the slots and slot configurations of the business to be handled, specifically including:

[0018] Receive the basic information voice of the business to be handled input by the user in contrast to the basic information guidance template;

[0019] Determine the business scenario of the business to be handled according to the basic information guidance template compared by the user;

[0020] Obtain the first slot, the second slot, and the third slot to be collected and filled and their groups according to the business scenario of the business to be handled.

[0021] Further, based on the slots and slot configurations, have a conversation with the user based on the large language model to complete the filling of slot information, specifically including:

[0022] Use the large language model to extract the first keyword in the basic information voice of the business to be handled, and fill the first slot with the first keyword;

[0023] Use the large language model to sequentially propose user questions for each group of second slots to be collected and filled, obtain the user answers according to the user questions, extract the second keyword in the user answers, and fill the second slot with the second keyword;

[0024] Sequentially display or broadcast the component cards of each group of third slots to be collected and filled, obtain the card content selected by the user according to the component cards, and fill the third slot according to the card content selected by the user.

[0025] Further, the method further includes the following steps executed by the business order transfer system:

[0026] Convert the extracted first keyword, second keyword, and the content of the card selected by the user into the same format that meets the requirements of the business acceptance message through a large language model, and perform content verification;

[0027] Identify the voice of giving up handling the business, the voice of jumping to handle the business, or the voice of modifying slot information input by the user through a large language model;

[0028] Evaluate the accuracy of the prompt words for the user's question based on the passing rate of content verification, the abandonment rate of handling the business, the jumping rate of handling the business, or the slot modification rate for handling the business multiple times;

[0029] Dynamically update the prompt words based on the accuracy of the prompt words in combination with the professional knowledge base.

[0030] Further, fuse the information filled in all slots of the business to be handled to generate a business acceptance message, and send the business acceptance message to the business acceptance system, specifically including:

[0031] Obtain the slot information in the same format in all slots of the business to be handled, and fill it into the corresponding business acceptance message template to generate a business acceptance message;

[0032] Display the business acceptance message and / or the list of slot information to the user for inspection through a preview component, and receive the submission instruction sent by the user after checking and confirming;

[0033] Obtain the application programming interface (API) corresponding to the business scenario of the business to be handled, send the business acceptance message to the business acceptance system through the corresponding API according to the submission instruction, and receive the business acceptance order number returned by the business acceptance system.

[0034] Further, among them:

[0035] The business is a government and enterprise business, the specific form of the business order transfer system is an application APP used by a government and enterprise business customer manager or an application APP playing the role of a government and enterprise business customer manager, the user is a government and enterprise business customer manager and / or a government and enterprise business customer, the business scenarios include basic government and enterprise business, dual-line business, and innovative business, the professional knowledge base includes government and enterprise middle platform operation manuals, intelligent acceptance operation manuals, and business acceptance specifications, and both the slot information and the business acceptance message adopt the JSON format.

[0036] In a second aspect, the present application provides a business order transfer acceptance device, the device is specifically a business order transfer system, or a business order transfer acceptance system including a business acceptance system and a business order transfer system, and the business order transfer system includes:

[0037] A slot module, used to obtain the slots and slot configurations of the business to be processed;

[0038] A dialogue module, connected to the slot module, used to dialogue with the user based on the large language model according to the slots and slot configurations to complete the filling of slot information;

[0039] A message module, connected to the dialogue module, used to fuse the information filled in all the slots of the business to be processed to generate a business acceptance message and send the business acceptance message to the business acceptance system.

[0040] Thirdly, the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is run by a processor, the above-mentioned business order transfer acceptance method is implemented.

[0041] The present application provides a business order transfer acceptance method, device and medium. By designing the slots in the process of handling business order transfer acceptance, using the large language model to dialogue with the user based on the slots, guiding the user to provide the necessary information for business acceptance, and automatically generating a business acceptance message through slot filling, the intelligent collection of business acceptance information and the efficient, accurate and automated business processing process are realized. Brief Description of the Drawings

[0042] Figure 1 is a flowchart of a business order transfer acceptance method according to an embodiment of the present application;

[0043] Figure 2 is a schematic structural diagram of a business order transfer acceptance system according to an embodiment of the present application;

[0044] Figure 3 is a schematic structural diagram of a business order transfer system according to an embodiment of the present application;

[0045] Figure 4 is an example diagram of slot definition configuration according to an embodiment of the present application;

[0046] Figure 5 is an example diagram of a component card according to an embodiment of the present application;

[0047] Figure 6 is a schematic diagram of dialogue with the user based on the large language model according to an embodiment of the present application. Detailed Embodiment

[0048] To enable those skilled in the art to better understand the technical solutions of the present application, the following will further describe the embodiments of the present application in detail with reference to the drawings.

[0049] It can be understood that the specific embodiments and drawings described herein are only used to explain the present application, rather than limiting the present application.

[0050] It is understood that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other.

[0051] It is understood that, for ease of description, only the parts related to the present application are shown in the drawings of the present application, and the parts unrelated to the present application are not shown in the drawings.

[0052] It is understood that each module and unit involved in the embodiments of the present application may correspond to only one entity structure, or may be composed of multiple entity structures. Alternatively, multiple modules and units may also be integrated into one entity structure.

[0053] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of the present application may occur in a different order from that marked in the drawings.

[0054] It is understood that in the flowcharts and block diagrams of the present application, the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the embodiments of the present application are shown. Among them, each block in the flowchart or block diagram may represent a module, unit, program segment, or code, which contains executable instructions for implementing the specified function. Moreover, each block or combination of blocks in the block diagram and flowchart may be implemented by a hardware-based device for implementing the specified function, or may be implemented by a combination of hardware and computer instructions.

[0055] It is understood that the modules and units involved in the embodiments of the present application may be implemented in software or in hardware. For example, the modules and units may be located in the processor.

[0056] Embodiment 1:

[0057] As Figure 1 shown, the present application provides a method for accepting transferred business orders. The method is applied to a business order transfer system 1 as Figure 2 shown, or a business order transfer acceptance system including a business acceptance system 2 and a business order transfer system 1. The method at least includes the following steps executed by the business order transfer system 1:

[0058] S11. Obtain the slots and slot configurations of the business to be processed;

[0059] S12. Based on the slots and slot configurations, conduct a conversation with the user based on a large language model to complete the filling of slot information;

[0060] S13. Integrate the information filled in all the slots of the business to be processed to generate a business acceptance message, and send the business acceptance message to the business acceptance system 2.

[0061] In this embodiment, the method designs slots in the process of accepting dropped orders for handling services, uses a large language model to communicate with the user based on the slots, guides the user to provide necessary information for service acceptance, and automatically generates a service acceptance message through slot filling, realizing intelligent collection of service acceptance information and an efficient, accurate and automated service processing flow. As Figure 1 The method shown is applied to the device shown in Figure 2 and 3 shown.

[0062] Specifically, this embodiment provides a method for intelligent acceptance of operator's government and enterprise services. In order to simplify the acceptance process of government and enterprise services and improve the processing speed, this method deeply integrates open-source large models and artificial intelligence technologies, and cooperates with natural language processing (NLP, Natural Language Processing) and speech recognition technologies. Through the application of these technologies, key elements for service acceptance are accurately extracted from the conversation information provided by the user. When the information is incomplete or more detailed information is required, this method will guide the user to supplement necessary details in the form of interactive cards to ensure that all necessary information slots are fully filled. The application of this method greatly reduces the difficulty for business personnel to learn new services, reduces the time consumption of manual entry, and significantly improves the overall work efficiency and service quality.

[0063] In one implementation, before S11, obtaining the slots and slot configurations of the service to be handled, the method further includes the business order dropping system 1 performing the following steps:

[0064] Pre-design service acceptance message templates corresponding to multiple service scenarios, each service acceptance message template having service elements to be collected and filled;

[0065] Taking each service element to be collected and filled as a slot, and configuring the collection and filling rules for each slot information;

[0066] Grouping the slots according to the service scenarios and collection and filling rules, so that the slots in the same group can complete information collection at one time and be filled into the same service acceptance message template.

[0067] In this embodiment, the core of the method is to build an intelligent dialogue processing system that can automatically identify the user's intention and extract key business information, organize this information in the form of slots, and finally trigger the business acceptance process after the slots are filled. Through slot design in NLP interaction, the key information in the dialogue can be captured in a timely manner and made easy to structure, facilitating the systematic verification of business elements during the subsequent business acceptance process. Based on each specific government and enterprise business scenario, all the slots required for this business scenario and the attributes corresponding to each slot are designed. The information possessed by each slot includes name, data type (such as text, number, date, object, component, etc.), whether it is required, default value, optional value, verification rule, prompt message, explanatory note, etc. That is, a lot of slots are predefined for government and enterprise business, and each slot collects a necessary piece of information for handling government and enterprise business. For a specific government and enterprise business scenario, multiple slots that collect the information required for this scenario are selected from them. For example Figure 4 as shown, an example of the definition configuration of one of the slots is given. The user or designer can specify what this slot is, what content needs to be filled, what filling requirements there are, etc. The example in the figure requires defining the business type and the configuration method of this business element. During actual operation, after the large model recognizes the key information, it converts and assigns the key information according to the slot definition.

[0068] Specifically, for government and enterprise business in this embodiment, it is defined which business scenarios are supported, the acceptance steps of a single business scenario are split, and the business elements required within each step (i.e., slots: name, description, encoding, value type, value range, whether it is required, etc.) are set, keeping the slot structure consistent. Slots are defined according to the business. For example, for the dual-line business, there may be more than 80 mandatory slots, including: customer code, customer name, business license certificate, customer contact person, customer contact phone number, product code, product name, product description, business number, communication address, verification business type, A-end address, A-end interface type, Z-end address, Z-end interface type, rate, A-end contact person, A-end contact phone number, Z-end contact person, Z-end contact phone number, monthly rent, first-month charging method, whether it is associated with a contract, contract, attachment for tariff approval, production mode, developer code, developer name, development channel, special instructions, etc. Examples of some of the designed slots (i.e., the definition of business elements in the message) are shown in Table 1 below:

[0069] Table 1 Example Table of Slot Design for Government and Enterprise Business Acceptance Message

[0070]

[0071]

[0072] An example of a message (service acceptance message) is as follows, which involves filling in some information corresponding to the slots in Table 1:

[0073]

[0074]

[0075] In one embodiment, the slots are grouped according to the service scenario and the collection and filling rules, so that the slots in the same group can complete information collection at one time and be filled into the same service acceptance message template. Specifically, it includes:

[0076] According to the collection and filling rules, the slots are at least divided into three categories: the first slot, the second slot, and the third slot. Each category of slots is divided into several groups according to the service acceptance message template involved;

[0077] According to the first slot, a basic information guidance template for the business to be handled is designed. The basic information guidance template is used to guide the user to input the basic information of the business to be handled and can at least identify the service scenario of the business to be handled;

[0078] According to each group of second slots, a prompt word for generating a user question based on the large language model is designed, so that the large language model generates a user question for collecting the information of each group of second slots according to the prompt word;

[0079] According to each group of third slots, a component card is designed. The component card is used to restrict the user to input the information to be filled into the third slot in a selected manner.

[0080] In this embodiment, the slots that need to be collected and filled in the message corresponding to each service scenario are classified and grouped; the classification is to divide the slots that can collect information in the same way into one category, including the first slot for collecting basic information when starting the service acceptance handover, the second slot for guiding the user to provide information through user Q&A, and the third slot for guiding the user to accurately fill in information. The first category of slots guides the user to input basic information through a fixed voice guidance template, the second category generates user questions intelligently through the large language model to guide the user to answer to obtain information, and the third category provides cards for the user to select; the purpose of grouping is to simplify the system design. According to whether the slot is applied in one or several message templates, combined with the user's thinking habit of answering questions, the question-and-answer ability of the large language model, etc., the slot grouping is designed to minimize the number of conversations and improve the efficiency of information collection.

[0081] Specifically, the interaction between the user and the system begins with voice input. To convert the user's voice into processable text information, Speech-to-Text (STT) technology is adopted. STT technology utilizes deep learning models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and their variants like Long Short-Term Memory (LSTM, a type of time-recurrent neural network) and Gated Recurrent Unit (GRU, an improved recurrent neural network) to accurately convert voice signals into text content. The user's basic information input in voice, such as "I want to handle XXX business" and the corresponding business information. Taking broadband as an example: "I want to open a ***M broadband for Zhang San at No. *** Road, *** City. The customer is in a hurry and expects it to be opened before ** / ** / ****." The user in this embodiment may be a customer manager of the operator's government and enterprise business, but it does not exclude directly providing the business order transfer system 1 provided in this embodiment to the customers of the operator's business. The converted text information is sent into a fine-tuned large language model (LLM). Before that, the LLM will be fine-tuned according to specific business scenarios with prompts, that is, by providing customized prompts or templates to guide the model to better understand the business background and requirements, thereby improving its ability to judge user intentions and extract business information.

[0082] In one embodiment, S11: Obtain the slots and slot configurations of the business to be handled, specifically including:

[0083] Receive the basic information voice of the business to be handled input by the user with reference to the basic information guidance template;

[0084] Determine the business scenario of the business to be handled according to the basic information guidance template referred to by the user;

[0085] Obtain the first slot, the second slot, the third slot to be collected and filled and their groupings according to the business scenario of the business to be handled.

[0086] In this embodiment, based on the open-source Tongyi Qianwen large model, intent recognition is implemented, and multi-turn dialogue capabilities are available. The core is to customize and manage the prompt, including: providing a prefabricated set of slot information, natural language templates, role-playing definitions, task descriptions, restrictive text, business acceptance knowledge bases, etc., dynamically generating the prompt for the current scenario, and combining it with the user's input, and submitting them to the large model for business recognition and information extraction and filling. For each business, 1-3 natural language templates for acceptance guidance are designed. For example, for the broadband business, it can be set as: Apply for *** Mbps broadband for Zhang San at *** Road, *** No., 186********. The customer service manager indicates the basic business handling requirements according to this template, and the system finds the slots that need to be filled according to this requirement, and then guides the customer service manager to collect the information of the unfilled slots. Currently, the system plays the role of a China Unicom customer service manager, responsible for tasks such as completing customer business acceptance and answering customer questions. Subsequently, the role definition and task service content can be added, deleted, modified according to the actual business acceptance situation.

[0087] In one embodiment, S12. Based on the slots and slot configurations, have a dialogue with the user using the large language model to complete the filling of slot information, specifically including:

[0088] Use the large language model to extract the first keyword in the voice of the basic information of the business to be handled, and use the first keyword to fill the first slot;

[0089] Use the large language model to sequentially ask the user questions for each group of second slots to be collected and filled, obtain the user's answer according to the user's question, extract the second keyword in the user's answer, and use the second keyword to fill the second slot;

[0090] Sequentially display or broadcast the component cards for each group of third slots to be collected and filled, obtain the card content selected by the user according to the component cards, and fill the third slot according to the card content selected by the user.

[0091] In this embodiment, currently two methods are used to collect slot information: if it can be directly collected from the dialogue, collect it from the dialogue, including collecting the information of the first slot and the second slot; if it cannot be collected from the dialogue, configure it into a component, after identifying it from the dialogue, answer it to the customer in the form of a card, and let the customer operate from the card, such as: select group customers, select business acceptance products, select developers, locate through GIS (Geographic Information System), etc. To ensure that business requirements may change over time, each slot supports dynamic update. At the same time, to ensure the user experience and ensure that the guiding information is clear and concise, the component cards can all ensure that the information collection is completed within 1-2 steps of operation. An example of a component card is Figure 5As shown, in a conversation with a user, a component is a card, and a component may contain multiple slots. Within a component, through manual active operations, multiple slots can be directly filled. By combining the requirements for collecting information from some slots through components, the collection and filling of slot information can be completed at one time, improving the collection efficiency. At the same time, an interaction channel with the business system is also provided, such as business number verification, product selection, ID card recognition, etc.

[0092] In one embodiment, the method further includes the following steps performed by the business order transfer system:

[0093] Convert the extracted first keyword, second keyword, and the content of the card selected by the user into the same format that meets the requirements of the business acceptance message through a large language model, and perform content verification;

[0094] Identify the voice of the user giving up handling the business, or the voice of jumping to handle the business, or the voice of modifying slot information input by the user through a large language model;

[0095] Evaluate the accuracy of the prompt words for the user's question according to the content verification passing rate, business abandonment rate, business jump rate, or slot modification rate of handling the business multiple times;

[0096] Dynamically update the prompt words according to the accuracy of the prompt words in combination with the professional knowledge base.

[0097] In this embodiment, as Figure 6 shown, based on the large language model LLM, multi-round dialogue interaction with the user can be realized. In the figure, the constituent elements may specifically include: intent recognition, intent determination, conversation management, slot extraction, slot fusion, construction of prompt words, reply content, etc. For the prompt (prompt words), an independent management function is provided, including: module management, evaluation management, and log management functions.

[0098] Module management classifies different modules according to business scenarios and functions, and uses different prompt word templates, including providing the prompt words used by the model to extract the words for filling the slots. The current prompt word templates include: intent recognition prompt word template, business recognition prompt word template, business modification prompt word template, etc. User business intent recognition prompt word template: According to the current environment of the user and the input language, determine whether the user wants to handle the business, what kind of business to handle, and whether to give up handling the business, etc. Business information recognition prompt word template: Identify the key information from the user's voice conversation and correspond it to the business slot information. Slot information modification prompt word template: Identify whether the user has a modification intention from the user's voice input. If there is a modification intention, identify the key information and compare it with the filled slot information for coverage.

[0099] Log management: Persistently record the generated prompts and synchronously record the LLM inputs and outputs. Evaluation management: Compare the LLM inputs and outputs, as well as the slot information extraction, as the basis for evaluating the quality control of the prompts, so as to continuously optimize and update the prompts in the later stage. The current comparison basis mainly includes: the number of multi-round conversations, the business cancellation rate, and finally evaluate the prompt quality through the passing rate of slot information verification, etc. This independent prompt management implementation serves as a communication bridge between business acceptance and the interaction with the large model, enabling real-time monitoring of the large model's recognition accuracy, dynamically updating and adjusting the prompt words, and improving the accuracy of the large model in extracting keywords.

[0100] Multi-round conversation process: After the user inputs through text or voice, it will first go through the system for user intent recognition. If the intent recognition determines that this is a new business or the user's business handling intent remains the same (i.e., there is no jump), it will enter the slot extraction process. The model will fuse the extracted key information into slot filling. At the same time, according to the pre-configured prompt word template, combined with the business knowledge base information, construct the prompt words, and finally construct the content to reply to the user in combination with the user input information, guiding the user to fill in the information of other slots. If the system determines that the user has an intent jump during intent recognition, the original conversation will be incorporated into the conversation management and recorded as a historical conversation. At the same time, prompt the user whether to start a new round of conversation. When starting a new round of conversation, temporarily store the original conversation, and the customer manager can switch back from the conversation list at any time.

[0101] Finally, the large model is responsible for parsing the user's conversation, identifying the user's intent, and extracting the necessary business element information from the conversation. These information are classified and filled according to the predefined slot structure. The slot design covers all key parameters that may affect business acceptance, such as customer information, product information, etc. When all the necessary slots are correctly filled, a corresponding business acceptance message will be automatically generated and submitted to the business acceptance system through the API (Application Programming Interface) interface to complete the acceptance process.

[0102] It is agreed that the response results of the large language model (LLM) will be in JSON (JavaScript Object Notation) format. Based on the multi-round conversation process, in combination with the historical conversation records, slot information, and the generated prompt, the LLM helps extract the information required for each slot, writes the successfully extracted content as a character type into the value under the corresponding slot key, and finally integrates it as a whole and outputs it in JSON format for subsequent generation of business acceptance messages. Given the independent management ability provided for the prompt, the current information extraction ability already supports extracting multiple slot information at one time. The extraction ability depends on the computing power and token limit of the LLM. At the same time, it also supports modifying the information that has completed the slot fusion verification. The extracted content is the business information required in the business acceptance process, and the key business information is extracted according to the slot definitions such as Figure 1 shown. After extraction, verification will be carried out first. After passing the verification, it will be organized into a business message (refer to the message sample). The business message gradually increases. Token is the word limit in the LLM session. The larger the text quantity, the more words are required. For example: Handle *** megabit broadband for Zhang San at *** Road, *** No., 186********. Multiple information such as customer name, product, address, and contact phone number can be extracted at one time. Content verification may include: performing sequential verification in the defined order, verifying whether it has been filled, whether it conforms to the defined type, etc. If components are set, the components will provide independent verification logic, which can provide more complex business logic verification, such as whether the business number is normal, whether the product code is true and valid, whether there are installation resources at the installation address, etc.

[0103] A sample of the LLM session interface is as follows:

[0104]

[0105]

[0106] The conversation process shown in this sample is: The account manager actively says that they want to handle a broadband account opening business. The system organizes the prompt and guiding language and responds to the account manager. According to the guiding language, the account manager then replies with the installation address.

[0107] Description of parameters for the large model conversation interface: model (string, required): the ID of the model to be used; messages (array, required): the list of messages describing the conversation; role (string, required): the role sending this message, one of system (system, for reply content), user (user, representing the China Unicom customer manager, for asking questions), or assistant (assistant, for storing prompt words to facilitate better inference by the model). Generally, user is used to send user questions and system is used to send prompt information to the model; content (string, required): the content of the message; name (string, optional): the name of the sender of this message, which can contain a-z, A-Z, 0-9, and underscores, with a maximum length of 64 characters; stream (boolean, optional, whether to send content in a streaming manner): when it is set to true, the API will return content in the SSE (Server-Side Event) manner. If it is not for real-time chat, the default is false; max_tokens (integer, optional): the maximum number of tokens generated in chat completion. The total length of input tokens and generated tokens is limited by the model context length; temperature (number, optional, default is 1): the sampling temperature, between 0 and 2. Higher values are more random, while lower values are more focused and deterministic.

[0108] Slot filling process: According to the parsing output result of the large model, perform rule verification according to the rules predefined for the corresponding slots. After successful verification, automatically fill the corresponding slots, and at the same time check whether any of the next required slots are empty or do not meet the verification rules. The predefined rules include: verifying according to the value type (string, number, enum value, date, array, object, etc.); if a component is defined, the component provides the actual verification logic for slot verification to call. If the actual content fails the verification, it is all handled as if the content is empty.

[0109] Organize the reply content: If there are required slots that are empty or do not meet the verification rules, the system organizes the reply content according to the predefined natural language template. The response example is as follows: Please enter the following information: customer contact person, customer contact phone number; if the required information belongs to precisely customized content (such as: select customer, select product, address location selection, etc.), an operation card (component-style card) will be popped up synchronously to facilitate further information collection. The component configuration management function is provided in the business configuration. After the slot is defined, a component can be created to incorporate business elements into the component.

[0110] If all slots are collected completely, it will finally enter the process of previewing and submitting the message organization.

[0111] In one embodiment, S13 generates a service acceptance message by integrating the information filled in all the slots of the service to be processed, and sends the service acceptance message to the service acceptance system, which specifically includes:

[0112] Obtain the slot information in the same format in all the slots of the service to be processed, and fill it into the corresponding service acceptance message template to generate a service acceptance message;

[0113] Display the service acceptance message and / or the slot information list to the user for inspection through a preview component, and receive the submission instruction sent by the user after the inspection is correct;

[0114] Obtain the application programming interface (API) corresponding to the service scenario of the service to be processed, send the service acceptance message to the service acceptance system through the corresponding API according to the submission instruction, and receive the service acceptance order number returned by the service acceptance system.

[0115] In this embodiment, after the slot configuration is completed for each service acceptance scenario, there is a final submission API interface and a final submission message sample (message template, used to automatically and intelligently organize the service message), that is, the message submitted by the service order-offloading system to the service acceptance system. A corresponding message sample is pre-designed for a service scenario, and the content included is defined according to the actual service. After the slot information is collected, this method will automatically assemble the message automatically according to this sample. After the assembly is successful, it will be handed over to the system preview component for display to the user. The user actively confirms the service acceptance information, allows direct modification of some slot information (the slot attribute is set to allow direct modification), and actively initiates the service acceptance submission function. After the system receives the user's submission request, it actively initiates a call to the API interface of this service scenario, completes the submission of the service acceptance message, and responds to the user with the corresponding acceptance order number.

[0116] A slot is equivalent to an element in the message. The assembly process is to design a JSON. According to the defined order of the slots, every time one passes the verification, a corresponding information is added to the JSON until all the slots pass the verification and are filled. At this time, the JSON is the message to be finally submitted. Generally, for the slot information processed by components (cards), there are enumerated values and linkage relationships between elements, and they cannot be modified. API interfaces such as: there is a corresponding broadband submission interface for broadband acceptance, and a dual-line service acceptance account opening interface for dual-line service acceptance, etc.

[0117] Another message sample is as follows:

[0118]

[0119] In the above message sample, the meaning of each key is user-defined and is the final effect of the definition of some keys in Table 1.

[0120] In one embodiment, where:

[0121] The business is government and enterprise business. The specific form of the business order transfer system is an application APP used by government and enterprise business customer managers or an application APP playing the role of a government and enterprise business customer manager. The users are government and enterprise business customer managers and / or government and enterprise business customers. The business scenarios include basic business, dual-line business, and innovative business of government and enterprise business. The professional knowledge base includes operation manuals of the government and enterprise middle platform, intelligent acceptance operation manuals, and business acceptance specifications. Both the slot information and the business acceptance message are in JSON format.

[0122] In this embodiment, the business scenarios included in the government and enterprise intelligent acceptance (order transfer) system are: basic business, dual-line business, innovative business, etc.; the content of the knowledge base mainly includes: operation manuals of the government and enterprise middle platform, intelligent acceptance operation manuals, business acceptance specifications, etc., and its function is to enrich the prompt words; a possible specific implementation form is that the customer manager handles the government and enterprise order transfer business within the APP (application, such as an application on a mobile phone), and the system responds with an order transfer number to the customer manager.

[0123] This embodiment proposes the concept design of slots in business acceptance, designs the business order transfer acceptance process based on the intention recognition of the large model and multi-round dialogue, and also proposes the management function design of the large model prompt; it realizes the government and enterprise business acceptance function of the operator based on the open-source large model. This method constructs an intelligent dialogue processing system, converts the user's speech into text through the speech-to-text (STT) technology, and then the fine-tuned large language model (LLM) parses the dialogue, recognizes the intention and extracts key business information in a specific business scenario. These information are organized in the form of slots, covering all key parameters affecting business acceptance. After the slots are filled, the system automatically generates a business acceptance message and submits it to the business system through the API interface to complete the acceptance, realizing an efficient and accurate automated business processing process.

[0124] Embodiment 2:

[0125] As Figure 2 and 3 shown, this application provides a business order transfer acceptance device, which is specifically the business order transfer system 1, or a business order transfer acceptance system including the business acceptance system 2 and the business order transfer system 1. The business order transfer system 1 includes:

[0126] A slot module 11, used to obtain the slots and slot configurations of the business to be handled;

[0127] A dialogue module 12, connected to the slot module 11, used to have a dialogue with the user based on the large language model according to the slots and slot configurations to complete the filling of slot information;

[0128] The message module 13 is connected to the dialogue module 12 and is used to generate a business acceptance message by integrating the information filled in all the slots of the business to be processed, and send the business acceptance message to the business acceptance system.

[0129] In an embodiment, the business transfer system further includes a design module, which is connected to the slot module 11 and specifically includes:

[0130] The message design unit is used to pre-design business acceptance message templates corresponding to multiple business scenarios, and each business acceptance message template has business elements to be collected and filled.

[0131] The slot configuration unit is connected to the message design unit and is used to take each business element to be collected and filled as a slot, and configure the collection and filling rules for each slot information.

[0132] The slot grouping unit is connected to the slot configuration unit and is used to group the slots according to the business scenario and the collection and filling rules, so that the slots in the same group can complete information collection at one time and be filled into the same business acceptance message template.

[0133] In an embodiment, the slot grouping unit specifically includes:

[0134] The classification grouping unit is used to classify the slots into at least three categories: the first slot, the second slot, and the third slot according to the collection and filling rules, and divide each category of slots into several groups according to the business acceptance message template involved.

[0135] The first slot unit is connected to the classification grouping unit and is used to design a basic information guidance template for the business to be processed according to the first slot. The basic information guidance template is used to guide the user to input the basic information of the business to be processed and can at least identify the business scenario of the business to be processed.

[0136] The second slot unit is connected to the classification grouping unit and is used to design a prompt word for generating a user question based on the large language model according to each group of second slots, so that the large language model generates a user question for collecting the information of each group of second slots according to the prompt word.

[0137] The third slot unit is connected to the classification grouping unit and is used to design a component card according to each group of third slots. The component card is used to limit the user to input the information to be filled into the third slot in a selected manner.

[0138] In an embodiment, the slot module 11 specifically includes:

[0139] The voice receiving unit is used to receive the basic information voice of the business to be processed input by the user in contrast to the basic information guidance template.

[0140] A scenario determination unit, connected to the voice reception unit, is configured to determine the business scenario of the business to be handled according to the basic information guidance template corresponding to the user;

[0141] A slot acquisition unit, connected to the scenario determination unit, is configured to acquire the first slot, the second slot, the third slot to be collected and filled, and their groupings according to the business scenario of the business to be handled.

[0142] In one embodiment, the dialogue module 12 specifically includes:

[0143] A first collection and filling unit, configured to extract the first keyword in the basic information voice of the business to be handled using a large language model, and fill the first slot with the first keyword;

[0144] A second collection and filling unit, configured to sequentially ask user questions for each group of second slots to be collected and filled using a large language model, obtain user answers according to the user questions, extract the second keyword in the user answers, and fill the second slot with the second keyword;

[0145] A third collection and filling unit, configured to sequentially display or broadcast the component cards of each group of third slots to be collected and filled, obtain the card content selected by the user according to the component cards, and fill the third slot according to the card content selected by the user.

[0146] In one embodiment, the business transfer system further includes:

[0147] A conversion and verification unit, configured to convert the extracted first keyword, second keyword, and the card content selected by the user into the same format that meets the requirements of the business acceptance message using a large language model, and perform content verification;

[0148] A voice recognition unit, configured to recognize the voice of giving up handling the business, or the voice of jumping to handle the business, or the voice of modifying slot information input by the user using a large language model;

[0149] An evaluation unit, connected to the conversion and verification unit and the voice recognition unit, is configured to evaluate the accuracy of the prompt words of the user questions according to the content verification passing rate, the business handling abandonment rate, the business handling jump rate, or the slot modification rate of handling the business multiple times;

[0150] An update unit, connected to the evaluation unit, is configured to dynamically update the prompt words according to the accuracy of the prompt words in combination with the professional knowledge base.

[0151] In one embodiment, the message module 13 specifically includes:

[0152] A message generation unit, configured to acquire the slot information in the same format in all slots of the business to be handled, and fill it into the corresponding business acceptance message template to generate a business acceptance message;

[0153] A message checking unit, connected to the message generating unit, is configured to display the service acceptance message and / or the slot information list to the user for checking through a preview component, and receive the submission instruction issued by the user after the user checks and confirms that there is no error.

[0154] A message sending unit, connected to the message checking unit, is configured to obtain the application programming interface (API) corresponding to the service scenario of the business to be processed, and send the service acceptance message to the service acceptance system through the corresponding API according to the submission instruction, and receive the service acceptance order number returned by the service acceptance system.

[0155] In one embodiment, where:

[0156] The service is an enterprise service. The specific form of the service order transfer system is an application APP used by an enterprise service account manager or an application APP playing the role of an enterprise service account manager. The user is an enterprise service account manager and / or an enterprise service customer. The service scenarios include basic services, dual-line services, and innovative services of enterprise services. The professional knowledge base includes the operation manual of the enterprise middle platform, the intelligent acceptance operation manual, and the service acceptance specification. Both the slot information and the service acceptance message are in JSON format.

[0157] Example 3:

[0158] Embodiment 3 of the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is run by a processor, it implements the service order transfer acceptance method as described in Embodiment 1, or implements the service order transfer acceptance device as described in Embodiment 2.

[0159] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information such as computer-readable instructions, data structures, computer program units, or other data. The computer-readable storage medium includes, but is not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), digital versatile disc (DVD) or other optical disc storage, magnetic cassette, tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0160] In addition, the present application can also provide a computer device, including a memory and a processor. A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the business order transfer acceptance method as described in Embodiment 1. This computer device can be the business order transfer acceptance device as described in Embodiment 2.

[0161] Among them, the memory is connected to the processor. The memory can be a flash memory, a read-only memory, or other memories. The processor can be a central processing unit or a single-chip microcomputer.

[0162] Embodiments 1-3 of the present application provide a business order transfer acceptance method, device, and medium. By designing the slots in the process of handling business order transfer acceptance, using a large language model to communicate with the user based on the slots, guiding the user to provide the necessary information for business acceptance, and automatically generating a business acceptance message through slot filling, intelligent collection of business acceptance information and an efficient, accurate, and automated business processing process are realized.

[0163] It can be understood that the above embodiments are merely exemplary embodiments adopted to illustrate the principle of the present application. However, the present application is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present application, and these modifications and improvements are also regarded as the protection scope of the present application.

Claims

1. A method for accepting business orders, characterized in that: The method is applied to a business order rejection system, or a business order rejection acceptance system including a business acceptance system and the business order rejection system, and the method at least includes the following steps performed by the business order rejection system: Get the slot and slot configuration of the business to be processed; Communicate with the user based on the slot and slot configuration based on the large language model to complete the slot information filling; The information filled in all the slots for the pending business is integrated to generate a business acceptance message, and the business acceptance message is sent to the business acceptance system.

2. The method according to claim 1, characterized in that Before obtaining the slot and slot configuration of the service to be processed, the method further includes the following steps performed by the service order rejection system: Pre-design business acceptance message templates corresponding to various business scenarios, each of which contains business elements to be collected and filled; Each business element to be collected and filled is a slot, and the collection and filling rules of each slot information are configured; The slots are grouped according to the business scenarios and collection and filling rules so that the slots in the same group can complete information collection at one time and fill in the same business acceptance message template.

3. The method according to claim 2, characterized in that Slots are grouped according to business scenarios and collection and filling rules so that slots in the same group can complete information collection at one time and fill in the same business acceptance message template, including: According to the collection and filling rules, the slots are divided into at least three categories: the first slot, the second slot and the third slot, and each category of slots is divided into several groups according to the service acceptance message template involved; A basic information guidance template for the pending business is designed according to the first slot, where the basic information guidance template is used to guide the user to enter basic information of the pending business and at least be able to identify the business scenario of the pending business; According to each group of second slots, a prompt word for generating a user question based on the large language model is designed, so that the large language model generates a user question for collecting information of each group of second slots according to the prompt word; A component card is designed according to each group of third slots, and the component card is used to restrict the user to input information to be filled in the third slots in a selected manner.

4. The method according to claim 3, characterized in that Get the slots and slot configurations for pending services, including: Receive the basic information voice of the pending business input by the user according to the basic information guidance template; Determine the business scenario of the business to be handled based on the basic information guidance template compared by the user; The first slot, the second slot, and the third slot to be collected and filled and their groups are obtained according to the business scenario of the business to be handled.

5. The method according to claim 4, characterized in that According to the slot and slot configuration, the system communicates with the user based on the large language model to complete the slot information filling, including: Use the large language model to extract the first keyword in the basic information voice of the business to be handled, and use the first keyword to fill the first slot; Use the large language model to sequentially propose user questions for each group of second slots to be collected and filled, obtain user answers based on the user questions, extract second keywords from the user answers, and use the second keywords to fill the second slots; The component cards of each group of the third slots to be collected and filled are displayed or announced in sequence, the card content selected by the user is obtained according to the component cards, and the third slots are filled according to the card content selected by the user.

6. The method according to claim 5, characterized in that The method further comprises the following steps performed by the business order removal system: The extracted first keyword, second keyword and card content selected by the user are converted into the same format that meets the business acceptance message requirements through the large language model, and the content is verified; The large language model is used to recognize the user's voice input of giving up business processing, jumping to business processing, or modifying slot information; Evaluate the accuracy of prompt words for user questions based on the content verification pass rate, business abandonment rate, business jump rate, or slot modification rate of multiple business transactions; The prompt words are dynamically updated according to the accuracy of the prompt words and the professional knowledge base.

7. The method according to claim 6, characterized in that The information filled in all slots of the pending business is integrated to generate a business acceptance message, and the business acceptance message is sent to the business acceptance system, specifically including: Obtain slot information of the same format from all slots of the business to be processed, and fill it into the corresponding business acceptance message template to generate a business acceptance message; The business acceptance message and / or slot information list is displayed to the user for review through the preview component, and a submission instruction is received after the user has reviewed and confirmed that the information is correct. Obtain the application programming interface API corresponding to the business scenario of the business to be processed, send the business acceptance message to the business acceptance system through the corresponding API according to the submission instruction, and receive the business acceptance order number returned by the business acceptance system.

8. The method according to claim 7, characterized in that in: The business is government business, and the specific form of the business order system is an application APP used by a government business account manager or an application APP that plays the role of a government business account manager. The user is a government business account manager and / or a government business customer. The business scenarios include the basic business, dual-line business, and innovative business of government business. The professional knowledge base includes the government and enterprise middle office operation manual, the intelligent acceptance operation manual, and the business acceptance instructions. The slot information and business acceptance messages are both in JSON format.

9. A business order acceptance device, characterized in that: The device is specifically a business order rejection system, or a business order rejection acceptance system including a business acceptance system and a business order rejection system, and the business order rejection system includes: The slot module is used to obtain the slot and slot configuration of the business to be processed; The dialogue module is connected to the slot module and is used to communicate with the user based on the large language model according to the slot and slot configuration to complete the slot information filling; The message module is connected to the dialogue module and is used to integrate the information filled in all the slots of the pending business to generate a business acceptance message, and send the business acceptance message to the business acceptance system.

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 by the processor, the business order acceptance method according to any one of claims 1 to 8 is implemented.

Images