Task processing method and device based on large model, and storage medium

By using large models and virtual digital people on financial self-service equipment, the problem of low business processing efficiency on financial self-service equipment is solved, efficient and flexible business processing processes are achieved, and labor costs are reduced.

CN119003714AInactive Publication Date: 2024-11-22GRG BANKING EQUIPMENT CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411000123.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

On financial self-service equipment, users need bank staff to guide business processing, resulting in low efficiency and complex operation, especially in multi-user scenarios.

Method used

The task processing method based on the big model is adopted, and by obtaining user voice information, using the big model to conduct intention recognition and virtual digital human conversations, automatically guide users to complete business processing, including knowledge Q&A and business processing processes, reducing labor costs and improving efficiency.

Benefits of technology

It realizes efficient business handling without manual guidance on financial self-service equipment. It is suitable for complex task scenarios, has strong flexibility, and is adapted to real-time business updates and expansions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119003714A_ABST
    Figure CN119003714A_ABST
Patent Text Reader

Abstract

The invention discloses a task processing method and device based on a large model and a storage medium, and belongs to the technical field of artificial intelligence. The task processing method based on the large model comprises the steps that voice information input by a user is acquired; performing intention recognition on the voice information by using the large model to determine a task type; according to the task type, utilizing a large model to control the virtual digital human to carry out conversation with the user so as to determine task demand information; the corresponding task is processed according to the task demand information and the large model, so that the user can be automatically guided to carry out business handling on the financial self-service equipment without arranging a worker to carry out task guidance or manually writing a business handling program, the labor cost is effectively reduced, and the business handling efficiency and flexibility are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application belongs to the field of artificial intelligence technology, and in particular, relates to a task processing method, device and storage medium based on a large model. Background Art

[0002] With the continuous advancement of technology and business development, the application of financial self-service equipment in banks is becoming more and more popular. Users can handle banking business by themselves on financial self-service equipment without going to the counter, which greatly facilitates people's lives.

[0003] At present, when users conduct banking business on financial self-service devices, they need bank staff to guide them to input instructions. However, one staff member can usually only guide one user at a time. Once multiple users seek help on site, the staff member needs to run back and forth between different users, or more staff members need to be arranged to provide guidance, which makes the entire business process complicated and inefficient. Summary of the invention

[0004] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application proposes a task processing method, device, storage medium, electronic device and computer program product based on a large model, which can automatically guide users to handle business on financial self-service devices with high business handling efficiency.

[0005] In a first aspect, the present application provides a task processing method based on a large model, comprising:

[0006] Get the voice information input by the user;

[0007] Using a large model to perform intent recognition on the voice information to determine the task type;

[0008] According to the task type, using the large model to control the virtual digital human to communicate with the user to determine task requirement information;

[0009] Process the corresponding tasks according to the task requirement information and the large model.

[0010] In some embodiments, when the task type is a knowledge question answering type, the task requirement information includes question information, and processing the corresponding task according to the task requirement information and the large model includes:

[0011] Retrieving knowledge matching the problem information in a preset knowledge database;

[0012] generating answer information using the large model based on the retrieved knowledge and prompt engineering;

[0013] The answer information is displayed to the user to conduct a knowledge question and answer session.

[0014] In some embodiments, when the task type is a business handling type, the task requirement information includes a filled business word slot, and according to the task type, using the large model to control the virtual digital human to communicate with the user to determine the task requirement information includes:

[0015] Determine multiple business word slots;

[0016] The large model is used to control the virtual digital person to have multiple rounds of dialogue with the user according to the business word slot, so as to guide the user to fill in the business word slot.

[0017] In some embodiments, using the macro model to control the virtual digital person to have a conversation with the user according to the business word slot to guide the user to fill in the business word slot includes:

[0018] Using the large model, determine the filling process of the current business word slot, the filling process includes user guidance instructions and the interface to be called;

[0019] According to the filling process, controlling the virtual digital person to have a dialogue with the user to guide the user to fill the current business word slot;

[0020] If the current business word slot is filled, the next business word slot is selected as the current business word slot, and the step of using the large model to determine the filling process of the current business word slot is returned to execution until all the business word slots are filled and the conversation is ended.

[0021] In some embodiments, processing the corresponding task according to the task requirement information and the large model includes:

[0022] The large model is used to handle corresponding services for the user according to the filled service word slots and the user database.

[0023] In some embodiments, the large model-based task processing method further includes:

[0024] Generate recommendation information using the big model according to the user information and the handled business information of the user;

[0025] The recommendation information is displayed to the user through the virtual digital person.

[0026] In some embodiments, the large model-based task processing method further includes:

[0027] Obtain industry knowledge sample sets, multi-task instruction sample sets and guided learning sample sets;

[0028] Pre-training the large model using the industry knowledge sample set;

[0029] Using the multi-task instruction sample set to perform fine-tuning training on the pre-trained large model;

[0030] The guided learning sample set is used to perform reinforcement learning training on the large model after fine-tuning training.

[0031] In some embodiments, the multi-task instruction sample set includes an intention recognition instruction sample set, a word slot filling instruction sample set, and a dialogue tracking instruction sample set, and fine-tuning the pre-trained large model using the multi-task instruction sample set includes:

[0032] Through a preset loss function, the pre-trained large model is subjected to multi-task joint training using the intention recognition instruction sample set, the word slot filling instruction sample set and the dialogue tracking instruction sample set. The preset loss function includes the loss function of the intention recognition task, the loss function of the word slot filling task and the loss function of the dialogue state tracking task.

[0033] In a second aspect, the present application provides a task processing device based on a large model, comprising:

[0034] An acquisition module is used to acquire the voice information input by the user;

[0035] A recognition module, used to use a large model to perform intent recognition on the voice information to determine the task type;

[0036] A dialogue module, used to control the virtual digital human to dialogue with the user using the large model according to the task type, so as to determine task requirement information;

[0037] An execution module is used to process corresponding tasks according to the task requirement information and the large model.

[0038] In some embodiments, when the task type is a knowledge question answering type, the task requirement information includes question information, and the execution module is specifically used to:

[0039] Retrieving knowledge matching the problem information in a preset knowledge database;

[0040] generating answer information using the large model based on the retrieved knowledge and prompt engineering;

[0041] The answer information is displayed to the user to conduct a knowledge question and answer session.

[0042] In some embodiments, when the task type is business handling, the business requirement information includes filled business word slots, and the dialogue module is specifically used to:

[0043] Determine multiple business word slots;

[0044] The large model is used to control the virtual digital person to have multiple rounds of dialogue with the user according to the business word slot, so as to guide the user to fill in the business word slot.

[0045] In some embodiments, the dialogue module is specifically used to:

[0046] Using the large model, determine the filling process of the current business word slot, the filling process includes user guidance instructions and the interface to be called;

[0047] According to the filling process, controlling the virtual digital person to have a dialogue with the user to guide the user to fill the current business word slot;

[0048] If the current business word slot is filled, the next business word slot is selected as the current business word slot, and the step of using the large model to determine the filling process of the current business word slot is returned to execution until all the business word slots are filled and the conversation is ended.

[0049] In some embodiments, the execution module is specifically used to:

[0050] The large model is used to handle corresponding services for the user according to the filled service word slots and the user database.

[0051] In some embodiments, the execution module is further configured to:

[0052] Generate recommendation information using the big model according to the user information and the handled business information of the user;

[0053] The recommendation information is displayed to the user through the virtual digital person.

[0054] In some embodiments, the large model-based task processing device further includes a training module for:

[0055] Obtain industry knowledge sample sets, multi-task instruction sample sets and guided learning sample sets;

[0056] Pre-training the large model using the industry knowledge sample set;

[0057] Using the multi-task instruction sample set to perform fine-tuning training on the pre-trained large model;

[0058] The guided learning sample set is used to perform reinforcement learning training on the large model after fine-tuning training.

[0059] In some embodiments, the multi-task instruction sample set includes an intention recognition instruction sample set, a word slot filling instruction sample set, and a dialogue tracking instruction sample set, and the training module is specifically used to:

[0060] Through a preset loss function, the pre-trained large model is subjected to multi-task joint training using the intention recognition instruction sample set, the word slot filling instruction sample set and the dialogue tracking instruction sample set. The preset loss function includes the loss function of the intention recognition task, the loss function of the word slot filling task and the loss function of the dialogue state tracking task.

[0061] In a third aspect, the present application provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the large model-based task processing methods described above.

[0062] In a fourth aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the large model-based task processing method described in any one of the above items is implemented.

[0063] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the large model-based task processing methods described above.

[0064] The big model-based task processing method, device, storage medium, electronic device and computer program product provided in the embodiments of the present application obtain voice information input by the user; use the big model to identify the intent of the voice information to determine the task type; according to the task type, use the big model to control the virtual digital human to communicate with the user to determine the task requirement information; process the corresponding task according to the task requirement information and the big model, that is, based on the big model and the virtual digital human, various user tasks are performed through dialogue, so that not only can the user be automatically guided to conduct business on the financial self-service equipment, without the need to arrange staff to conduct task guidance, but also without the need to manually write business processing procedures, which effectively reduces labor costs, improves business processing efficiency and flexibility, and is suitable for various complex task scenarios, and can better cope with real-time updates and expansions of business. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0066] Figure 1 It is a flowchart of a task processing method based on a large model provided in an embodiment of the present application;

[0067] Figure 2 is another flowchart of the task processing method based on a large model provided in an embodiment of the present application;

[0068] Figure 3 This is a schematic diagram of the display interface of a financial self-service device during a business handling process provided by an embodiment of the present application;

[0069] Figure 4 is a structural diagram of a task processing device based on a large model provided in an embodiment of the present application;

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

[0071] Figure 6 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0072] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.

[0073] At present, when users conduct banking business on financial self-service devices, they need bank staff to guide them to input instructions. However, one staff member can usually only guide one user at a time. Once multiple users seek help on site, the staff member needs to run back and forth between different users, or more staff members need to be arranged to provide guidance, which makes the entire business process complicated and inefficient.

[0074] With the development of computer technology and the advancement of artificial intelligence technology, intelligent human-computer interaction has become the mainstream of human-computer interaction at this stage. In order to realize the intelligent interaction of financial self-service equipment, a common way is to write code programs for business processing procedures for various businesses in advance and deploy them in financial self-service equipment to automatically guide customers to handle the required business on financial self-service equipment according to the set process. This intelligent interaction method requires developers to write business processing programs in advance. Once there are many types of business, the program writing will be more complicated. In addition, the implementation of this interaction method is completely based on the set program. It can only perform simple human-computer interaction with users, cannot provide personalized services, and cannot better respond to real-time updates and expansions of businesses. The user experience is poor.

[0075] To solve at least one of the above technical problems, an embodiment of the present application provides a task processing method, device, storage medium, electronic device and computer program product based on a large model.

[0076] See also Figure 1 , Figure 1 1 is a flow chart of a task processing method based on a large model provided in an embodiment of the present application. The task processing method based on a large model is applied to an electronic device, which can be implemented as a financial self-service device, and the financial self-service device is deployed in a financial scenario such as a bank. The task processing method based on a large model can include the following steps 101-104, wherein:

[0077] 101. Obtain the voice information input by the user;

[0078] 102. Use the large model to identify the intent of the voice information to determine the task type;

[0079] 103. According to the task type, the large model is used to control the virtual digital human to communicate with the user to determine the task requirement information;

[0080] 104. Process the corresponding task according to the task requirement information and the large model.

[0081] In the embodiment of the present application, by leveraging the powerful dialogue capabilities of the big model, the virtual digital human is controlled to dialogue with the user to obtain task requirement information, and then the powerful analysis, understanding and generation capabilities of the big model are used to process the corresponding tasks according to the task requirement information. That is, based on the big model and the virtual digital human, various user tasks are performed through dialogue. This not only automatically guides users to conduct business on financial self-service equipment, but also eliminates the need to arrange staff to conduct task guidance and manually write business processing procedures. It has low labor costs and high business processing efficiency, and is suitable for various complex task scenarios, with a wide range of applications and strong flexibility.

[0082] The above steps 101 - 104 will be further described in detail below.

[0083] 101. Obtain the voice information input by the user.

[0084] Among them, users can input information by communicating with virtual digital people. For example, when the financial self-service equipment detects that the user is approaching through its own sensor (such as an infrared sensor), the virtual digital person can be controlled to actively provide voice guidance. For example, the virtual digital person actively broadcasts to the user the services it can provide, such as business processing, problem consultation, etc., so that the user can say the service he needs.

[0085] 102. Use a large model to identify the intent of speech information to determine the task type.

[0086] Among them, the Large Model (also called the Foundation Model) refers to a machine learning model with a large number of parameters (usually more than one billion) and a complex structure, which can process massive amounts of data and complete various complex tasks, such as natural language processing, computer vision, speech recognition, etc. When determining the task type, the Large Model must first convert speech into text through speech recognition, and then perform intent recognition on the text based on natural language understanding capabilities.

[0087] The task types generally include knowledge question and answer, business handling and small talk. Knowledge question and answer tasks are mainly used to answer questions related to industry knowledge, such as loan policy consultation, monetary policy consultation, etc. Business handling tasks are mainly used to handle various financial services provided by banks, such as account balance inquiry, loan handling, etc. Small talk tasks are mainly used to answer questions unrelated to industry knowledge, such as weather consultation, social hot spot consultation, etc. The large model can be equipped with the ability to complete small talk tasks based on the learning of a large amount of general knowledge, and can be equipped with the ability to complete knowledge question and answer tasks and business handling tasks based on the learning of a large amount of industry knowledge.

[0088] 103. According to the task type, the large model is used to control the virtual digital human to communicate with the user to determine the task requirement information.

[0089] Among them, task requirement information is the key information required to perform the task. Different task types usually correspond to different task requirement information. The large model can control the virtual digital human to have at least one round of dialogue with the user based on the task requirement information to be collected. During the dialogue process, it is necessary to call the interfaces of various hardware devices in the financial self-service device, such as calling the display interface for text and / or picture display, calling the keyboard interface for text input, calling the microphone interface for voice collection, calling the speaker interface for voice playback, calling the camera interface for image collection, and so on.

[0090] Specifically, when controlling the conversation between the virtual digital human and the user, the big model needs to generate guiding instructions to inform the user what information is required of the user, and call various software and hardware interfaces to assist in completing the conversation, such as calling the external data query library interface to query the user's relevant data, calling the microphone interface to collect the user's voice, etc. Guiding instructions need to be generated based on the actual task requirement information, and guiding instructions can be displayed in a variety of ways, such as through text display, or through voice broadcast, or a combination of text and voice. These are all done by conducting targeted training on the big model in advance to enable the big model to have the corresponding capabilities, so that in the actual conversation with the user, the big model can analyze and decide independently without the user having to write a program to set it up.

[0091] For example, when the task type is knowledge question answering and small talk, the task requirement information includes question information (i.e., specific question content), that is, see Figure 2 , Figure 2 1 is another flowchart of the task processing method based on the large model provided in the embodiment of the present application. The above step 103 specifically includes:

[0092] 103A. According to the task type, the large model is used to control the virtual digital human to communicate with the user to determine the problem information.

[0093] Among them, the large model can generate a guiding sentence to guide the user to input specific question content, such as the guiding sentence "Please state the specific question you want to consult", and control the virtual digital person to display the guiding sentence to the user. For example, the virtual digital person uses the speaker of the financial self-service device to broadcast the guiding sentence in voice form. At the same time, the virtual digital person can also use the display screen to display the guiding sentence in text form.

[0094] For example, when the task type is business processing, the task requirement information includes basic information required for business processing, such as account information, date information, etc. These basic information can be collected by filling in word slots, that is, the task requirement information includes the filled business word slots. Figure 2 The above step 103 specifically includes the following steps 1031B and 1032B, wherein:

[0095] 1031B. Determine multiple business word slots;

[0096] 1032B. Using the large model, the virtual digital person is controlled to have multiple rounds of dialogue with the user according to the business word slot to guide the user to fill in the business word slot.

[0097] Among them, when the big model recognizes that the user's intention is to handle business, it can first provide the user with all the business types that can be handled. When the user selects a certain type of business, the big model determines the corresponding business word slot based on the business type, and guides the user to fill in the word slot by controlling the virtual digital person to communicate with the user, and then obtains the task requirement information. Specifically, the business types include but are not limited to personal business, debit card business, credit card business, account management, investment and financial management, loan business, corporate business, etc. Each business type can be displayed as an option on the display screen of the financial self-service device for users to choose. For example, the user selects the type of business they need to handle by clicking the virtual button or physical button on the display screen. Business word slots include but are not limited to business account, password, card information, date information, etc. Different business types usually correspond to different business word slots. The big model can be trained with a large amount of business process related content in advance to enable it to have the ability to extract business word slots for various businesses.

[0098] After the big model determines the business word slots, it needs to first determine the filling process based on the business word slots, and then control the virtual digital human to have multiple rounds of dialogue with the user according to the filling process, and complete the filling of each business word slot through dialogue to collect task requirement information. That is, the above step 1032B can specifically include:

[0099] Using the large model, determine the filling process of the current business word slot, which includes user guidance instructions and the interface to be called;

[0100] According to the filling process, the virtual digital person is controlled to have a dialogue with the user to guide the user to fill the current business word slot;

[0101] If the current business word slot is filled, the next business word slot is selected as the current business word slot, and the step of using the large model to determine the filling process of the current business word slot is returned to be executed until all business word slots are filled and the dialogue is ended.

[0102] The filling process includes the main execution steps involved in the business word slot filling process, which are completed by the big model and the user. For example, the big model needs to use its powerful capabilities to do some data analysis and data generation work, such as analyzing the word slot filling information input by the user, generating intermediate results or guiding instructions, etc. The user needs to input the filling information, such as personal account, password, date, etc.

[0103] In the process of word slot filling, in addition to the above-mentioned data analysis and data generation work, the big model also needs to automatically track and update the conversation status between the virtual digital person and the user according to the context of the conversation. For example, after the user completes the filling of the current business word slot, the big model needs to analyze whether to call the interface to obtain the intermediate result, or continue to generate guiding instructions to collect the filling information of the next word slot from the user, or end this round of conversation based on the filling content of the existing word slot.

[0104] At the same time, in the process of word slot filling, the large model also needs to select appropriate software and hardware interfaces for calling. For example, if you want to query the user database to obtain intermediate results, call the external user database query interface. If you want to guide the user to fill the slot, call the touch screen interface, speaker interface, etc. of the financial self-service device, and display guiding instructions to the user in the form of text, voice, etc. If you want to obtain the word slot filling information entered by the user, call the microphone interface or keyboard interface to collect input information in voice or text form. In addition, the large model also needs to flexibly select the interface according to the privacy of the information to be collected. For example, for the collection of user input information with strong privacy, such as the need to collect user passwords in the balance inquiry business, the large model calls the keyboard interface for users to input in text form. For the collection of user input information that does not involve user privacy, such as the water flow time period that needs to be collected in the water flow printing business, the large model calls the microphone interface for users to input in voice form.

[0105] For example, see Figure 3 , Figure 3 It is a schematic diagram of the display interface of the financial self-service device during the business handling process provided by the embodiment of the present application. When it is recognized that user A intends to handle business, and user A has been guided to say that the specific business content is "I want to check the account balance", the large model first recognizes the intention of the sentence "I want to check the account balance", and obtains that the business type that user A needs to handle is "balance inquiry". At this time, the large model first determines the business word slot corresponding to the business type, such as "bank account" and "query password", and determines the filling process of each business word slot, such as first filling the business word slot "bank account", and then filling the business word slot "query password", and for the filling of "bank account", it is necessary to first call the user data query interface to query all bank accounts of user A in the bank, and then provide these bank accounts to user A for selection in an option manner, and obtain and analyze the user's selection information to obtain the filling information of the business word slot "bank account".

[0106] For example Figure 3In the query, if it is found that user A has two accounts: account 1 and account 2, the big model needs to generate a guiding instruction "Please confirm the bank account: account 1, account 2" to guide the user to make a choice. And considering that bank accounts involve personal privacy, the virtual digital person can be controlled to only display the guiding instruction in text, rather than in the form of voice broadcast. Then, the keyboard interface, microphone interface, etc. are called to obtain the selection information input by the user, such as calling the microphone interface to collect the voice of user A "select the first one". Then, the big model analyzes and understands the user's input information, and determines that the filling information of the business word slot "bank account" is "account 1". After completing the filling of the business word slot "bank account", the big model needs to update the dialogue status in combination with the dialogue history. For example, if it is analyzed that the business word slot "query password" is left unfilled, a guiding instruction "Please enter the query password" is generated to guide the user to continue to fill the slot through dialogue. In addition, considering that passwords involve personal privacy, it is necessary to call the keyboard interface rather than the microphone interface to collect the password information entered by the user. For example, if the password is 123456, the entered password is compared and matched with the password stored in the user database. If the match is successful, the business word slot "query password" is filled in, and the big model updates the conversation status to end the conversation. If there is no match, the big model generates a guiding instruction "The password is incorrect, please re-enter the query password" to continue guiding the user to enter the password until the match is successful.

[0107] 104. Process the corresponding task according to the task requirement information and the large model.

[0108] In some embodiments, see Figure 2 When the task type is a knowledge question answering type, the task requirement information includes question information, and the above step 104 specifically includes the following steps 1041A-1043A, wherein:

[0109] 1041A. Retrieve knowledge matching the question information from a preset knowledge database;

[0110] 1042A. Generate answer information using the large model based on the retrieved knowledge and prompt engineering;

[0111] 1043A. Display the answer information to the user for question answering.

[0112] Among them, the knowledge database is a database of relevant industry knowledge created in advance. The big model can call the database query interface to query the matching knowledge based on the knowledge field and question content covered by the question information. After that, the prompt engineering is used to construct prompt information for the queried knowledge, and the prompt information is processed through the powerful understanding and generation capabilities of the big model to generate the final answer information, avoiding the process of manually constructing question and answer pairs, and effectively reducing labor costs.

[0113] Specifically, industry-related data can be collected through multiple channels, and industry expertise can be accumulated to create a knowledge database. These channels include web crawlers, public data, bank accumulation, and API interfaces, and the collected data includes both structured data and semi-structured and unstructured data. Data collected through these channels can be further processed and analyzed using data annotation, text mining and other technologies, roughly forming three types of business knowledge databases, product knowledge databases, and policy and regulatory knowledge databases.

[0114] Among them, the business knowledge database covers the process knowledge of financial business handling, including but not limited to loan handling, financial management recommendations, credit card installments, insurance handling, online payment, passbook inquiries, etc. The product knowledge database contains specific product knowledge in the financial industry, such as but not limited to savings accounts, financial management funds, product features, promotional activities, investment strategies, risk assessments, etc. The policy and regulatory knowledge database contains specific policy and regulatory knowledge in the financial industry, such as but not limited to asset management, risk auditing, credit policy, monetary policy, insurance policy, credit management, etc. By building these types of knowledge databases and combining them with the powerful language understanding and generation capabilities of large models, we can provide users with a variety of professional services and enhance user experience.

[0115] In some embodiments, when the task type is business handling, the task requirement information includes filled business word slots. In this case, the above step 104 specifically includes:

[0116] 104B. Using this large model, corresponding services are provided to users according to the filled business word slots and user database.

[0117] For example, for the above Figure 3 In the "Balance Inquiry" service, after successfully collecting the user's bank account and inquiry password information, the balance of the bank account can be queried in the user database and the queried balance result can be displayed to the user. For example, the queried balance is 3456.78 yuan to complete the "Balance Inquiry" service.

[0118] In addition, after the business is completed, the task processing method based on the big model may also include:

[0119] Generate recommendation information using the big model based on the user information and the business information that has been handled by the user;

[0120] The recommendation information is displayed to the user through the virtual digital person.

[0121] After completing the specific business, the big model can also recommend related businesses to the user. Figure 3In the example, after the user completes the "balance inquiry" service, the big model recommends the user to open a mobile banking service based on the user's information and the information of the services that have been completed. The generated recommendation is "You have not yet opened a mobile banking service. It is more convenient to check the balance on the mobile banking service. Do you need to open a mobile banking service?" For example, after the user completes the "statement printing" service, the big model recommends the loan service to the user. The generated recommendation is "Our bank recently launched a loan product. Are you interested in learning about it?", etc. Once the user makes a corresponding confirmation reply, the conversation intention between the virtual digital person and the user will change. This process is completely managed and tracked by the big model, avoiding the configuration of manual processes, realizing a human-computer interaction experience with thousands of faces, and improving the intelligence level of financial self-service equipment.

[0122] It should be noted that these capabilities of the large model require the creation of a large number of samples in advance for targeted training. That is, the task processing method based on the large model may also include:

[0123] Obtain industry knowledge sample sets, multi-task instruction sample sets and guided learning sample sets;

[0124] Use the industry knowledge sample set to pre-train the large model;

[0125] Using the multi-task instruction sample set to perform fine-tuning training on the pre-trained large model;

[0126] The guided learning sample set is used to perform reinforcement learning training on the large model after fine-tuning training.

[0127] Among them, pre-training is mainly used to inject financial industry knowledge into the big model. The industry knowledge sample set mainly includes financial information, product introductions, business processes, policies and regulations and other documents collected from various channels. Fine-tuning training is mainly used to shape the big model's ability to solve specific tasks, such as knowledge questions and answers, business processing and other specific types of tasks. Reinforcement learning training uses human-aligned reinforcement learning to train the big model, teaching human preferences to the model so that it can perform better in various task scenarios. Through these three stages of training, the big model can learn financial industry knowledge and have the ability to complete specific tasks. At the same time, it is more in line with user preferences when generating guidance instructions and replying to discourse, so that the big model has more professional and accurate understanding and expression capabilities.

[0128] In some embodiments, during the fine-tuning training stage, the multi-task instruction sample set includes an intent recognition instruction sample set, a slot filling instruction sample set, and a dialogue tracking instruction sample set. The intent recognition instruction sample set is used to perform training for intent recognition tasks, the slot filling instruction sample set is used to perform training for slot filling tasks, and the dialogue tracking instruction sample set is used to perform training for dialogue state tracking tasks.

[0129] Among them, the intent recognition task involves the intention recognition of personal business, debit card business, credit card business, account management, investment and financial management, loan business, corporate business and other business operations. The word slot filling task involves the recognition and filling of key business word slots such as account information, card information, and date information. The dialogue state tracking task involves automatically updating the dialogue state according to the context information and planning the next step, such as making an external interface call to obtain intermediate results or continuing to collect information from the user or ending the dialogue.

[0130] In some embodiments, these tasks are jointly trained in the large model. In this case, the above step of "fine-tuning the pre-trained large model using the multi-task instruction sample set" specifically includes:

[0131] Through the preset loss function, the pre-trained large model is subjected to multi-task joint training using the intention recognition instruction sample set, the word slot filling instruction sample set and the dialogue tracking instruction sample set. The preset loss function includes the loss function of the intention recognition task, the loss function of the word slot filling task and the loss function of the dialogue state tracking task.

[0132] Among them, the calculation formula of the loss value Instruct_Loss of the preset loss function is as follows:

[0133] Instruct_Loss = Loss Intent +Loss Slot +Loss DST ;

[0134] Among them, Loss Intent is the loss value of the loss function of the intent recognition task, Loss Slot is the loss value of the loss function for the slot filling task, Loss DST The loss value of the loss function of the dialogue state tracking task. By jointly training a large model for multiple tasks, the accuracy and generalization of the large model in handling various types of business can be improved, automatically meeting the expansion needs of the business.

[0135] Although the large model obtained through the above-mentioned fine-tuning training stage has a strong ability to generate dialogues and can answer questions raised by users well, it lacks the ability to ask questions autonomously, and its performance in the guidance scenario of financial self-service equipment needs to be enhanced. At the same time, financial business processing often involves the privacy information of users, and the large model is required to generate beneficial, harmless, and honest reply words during the dialogue process. Therefore, after the large model in the embodiment of the present application is fine-tuned, it will further use the method of reinforcement learning and artificial feedback to perform reinforcement learning training on the large model, so that the reply information and guidance information generated by the large model are more in line with human habits.

[0136] Among them, in the reinforcement learning training stage, the samples in the guided learning sample set include guidance instructions and manually annotated human feedback information. A reward model is first trained through the guided learning sample set, and then based on the reward model, the PPO (Proximal Policy Optimization) strategy is used to perform reinforcement learning training on the large model, so that the responses and guidance generated by the large model are more in line with human habits. For example, in a dialogue scenario for collecting user information, if it involves collecting user privacy information, the large model can use a reasonable dialogue method to guide users to input information, such as displaying "Please check whether the following information is correct on the machine" on the display screen in text form to guide users to operate the buttons on the machine (such as the "Confirm" or "Cancel" button) to reply, rather than speaking specific privacy content in voice.

[0137] From the above, it can be seen that the task processing method based on the big model provided in the embodiment of the present application obtains the voice information input by the user; uses the big model to identify the intent of the voice information to determine the task type; according to the task type, uses the big model to control the virtual digital human to communicate with the user to determine the task requirement information; processes the corresponding task according to the task requirement information and the big model, that is, based on the big model and the virtual digital human, various user tasks are performed through dialogue, so that it can not only automatically guide users to conduct business on financial self-service equipment, without the need to arrange staff to conduct task guidance, but also without the need to manually write business processing procedures, effectively reducing labor costs, improving business processing efficiency and flexibility, and is suitable for various complex task scenarios, and can better cope with real-time updates and expansions of business.

[0138] According to the method described in the above embodiment, the present application embodiment also provides a task processing device based on a large model, which is used to execute the steps in the above task processing method based on a large model. Figure 4 , Figure 4 200 is a schematic diagram of a task processing device based on a large model provided in an embodiment of the present application. The task processing device based on a large model 200 is applied to an electronic device, which can be implemented as a financial self-service device, and the financial self-service device is deployed in a financial scenario such as a bank. Specifically, the task processing device based on a large model 200 includes an acquisition module 201, an identification module 202, a dialogue module 203 and an execution module 204, wherein:

[0139] The acquisition module 201 is used to acquire the voice information input by the user;

[0140] Recognition module 202, used to use the large model to perform intent recognition on the voice information to determine the task type;

[0141] A dialogue module 203 is used to control the virtual digital human to dialogue with the user using the large model according to the task type to determine task requirement information;

[0142] The execution module 204 is used to process the corresponding task according to the task requirement information and the large model.

[0143] In some embodiments, when the task type is a knowledge question answering type, the task requirement information includes question information, and the execution module 204 is specifically used to:

[0144] Retrieve knowledge matching the question information from a preset knowledge database;

[0145] Based on the retrieved knowledge and prompt engineering, the large model is used to generate answer information;

[0146] The answer information is displayed to the user for question answering.

[0147] In some embodiments, when the task type is business handling, the task requirement information includes filled business word slots, and the dialogue module 203 is specifically used to:

[0148] Determine multiple business word slots;

[0149] By using the large model, the virtual digital person is controlled to have multiple rounds of dialogue with the user according to the business word slot, so as to guide the user to fill in the business word slot.

[0150] In some embodiments, the dialogue module 203 is specifically used to:

[0151] Using the large model, determine the filling process of the current business word slot, which includes user guidance instructions and the interface to be called;

[0152] According to the filling process, the virtual digital person is controlled to have a dialogue with the user to guide the user to fill the current business word slot;

[0153] If the current business word slot is filled, the next business word slot is selected as the current business word slot, and the steps of using the large model to determine the filling process of the current business word slot are returned to be executed until all business word slots are filled and the dialogue is ended.

[0154] In some embodiments, the execution module 204 is specifically used to:

[0155] Utilizing this large model, corresponding services are provided to users based on the filled service word slots and user database.

[0156] In some embodiments, the execution module 204 is further configured to:

[0157] Generate recommendation information using the big model based on the user information and the business information that has been handled by the user;

[0158] The recommendation information is displayed to the user through the virtual digital person.

[0159] In some embodiments, the large model-based task processing device further includes a training module for:

[0160] Obtain industry knowledge sample sets, multi-task instruction sample sets and guided learning sample sets;

[0161] Use the industry knowledge sample set to pre-train the large model;

[0162] The multi-task instruction sample set is used to fine-tune the pre-trained large model;

[0163] The guided learning sample set is used to perform reinforcement learning training on the large model after fine-tuning training.

[0164] In some embodiments, the multi-task instruction sample set includes an intention recognition instruction sample set, a word slot filling instruction sample set, and a dialogue tracking instruction sample set, and the training module is specifically used to:

[0165] Through the preset loss function, the pre-trained large model is used to perform multi-task joint training using the intention recognition instruction sample set, the word slot filling instruction sample set and the dialogue tracking instruction sample set. The preset loss function includes the loss function of the intention recognition task, the loss function of the word slot filling task and the loss function of the dialogue state tracking task.

[0166] It should be noted that the specific details of each module unit in the above-mentioned large model-based task processing device 200 have been described in detail in the embodiment of the above-mentioned large model-based task processing method, and will not be repeated here.

[0167] As can be seen from the above, the task processing device 200 based on the big model provided in the embodiment of the present application obtains the voice information input by the user through the acquisition module 201; the recognition module 202 uses the big model to perform intent recognition on the voice information to determine the task type; the dialogue module 203 uses the big model to control the virtual digital human to dialogue with the user according to the task type to determine the task requirement information; the execution module 204 processes the corresponding task according to the task requirement information and the big model, that is, based on the big model and the virtual digital human, various user tasks are executed through dialogue, so that it can not only automatically guide users to conduct business on financial self-service equipment, but also eliminate the need to arrange staff to conduct task guidance and manually write business processing procedures, effectively reducing labor costs, improving business processing efficiency and flexibility, and is suitable for various complex task scenarios, and can better cope with real-time updates and expansions of business.

[0168] In some embodiments, the task processing device based on the large model in the embodiment of the present application can be an electronic device, or a component in the electronic device, such as an integrated circuit or a chip. The electronic device can be a terminal, or it can be other devices other than the terminal. Exemplarily, the electronic device can be a mobile phone, a tablet computer, a laptop computer, a PDA, a vehicle-mounted electronic device, a mobile Internet device (Mobile Internet Device, MID), an augmented reality (augmentedreality, AR) / virtual reality (virtual reality, VR) device, a robot, a wearable device, an ultra-mobile personal computer (ultra-mobile personal computer, UMPC), a netbook or a personal digital assistant (personaldigital assistant, PDA), etc., and can also be a server, a network attached storage (Network AttachedStorage, NAS), a personal computer (personal computer, PC), a television (television, TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiment of the present application.

[0169] In some embodiments, Figure 5 As shown, an embodiment of the present application also provides an electronic device 300, including a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 301. When the program is executed by the processor 301, each process of the above-mentioned large model-based task processing method embodiment is implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0170] It should be noted that the electronic device 300 in the embodiment of the present application includes a mobile electronic device and a non-mobile electronic device.

[0171] Figure 6 A schematic diagram of the hardware structure of an electronic device 400 provided for implementing an embodiment of the present application.

[0172] The electronic device 400 includes but is not limited to: a radio frequency unit 401, a network module 402, an audio output unit 403, an input unit 404, a sensor 405, a display unit 406, a user input unit 407, an interface unit 408, a memory 409, a processor 410 and other components.

[0173] Those skilled in the art will appreciate that the electronic device 400 may also include a power source (such as a battery) for supplying power to each component, and the power source may be logically connected to the processor 410 through a power management system, thereby implementing functions such as managing charging, discharging, and power consumption management through the power management system. Figure 6 The electronic device structure shown in the figure does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently, which will not be described in detail here.

[0174] It should be understood that in the embodiment of the present application, the input unit 404 may include a graphics processor (Graphics Processing Unit, GPU) 4041 and a microphone 4042, and the graphics processor 4041 processes the image data of the static picture or video obtained by the image capture device (such as a camera) in the video capture mode or the image capture mode. The display unit 406 may include a display panel 4061, and the display panel 4061 may be configured in the form of a liquid crystal display, an organic light emitting diode, etc. The user input unit 407 includes a touch panel 4071 and at least one of other input devices 4072. The touch panel 4071 is also called a touch screen. The touch panel 4071 may include two parts: a touch detection device and a touch controller. Other input devices 4072 may include, but are not limited to, a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, and a joystick, which will not be repeated here.

[0175] The memory 409 can be used to store software programs and various data. The memory 409 may mainly include a first storage area for storing programs or instructions and a second storage area for storing data, wherein the first storage area may store an operating system, an application program or instructions required for at least one function (such as a sound playback function, an image playback function, etc.), etc. In addition, the memory 409 may include a volatile memory or a non-volatile memory, or the memory 409 may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDRSDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchronous link dynamic random access memory (SLDRAM) and a direct memory bus random access memory (DRRAM). The memory 409 in the embodiment of the present application includes but is not limited to these and any other suitable types of memory.

[0176] The processor 410 may include one or more processing units; the processor 410 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It is understandable that the modem processor may not be integrated into the processor 410.

[0177] An embodiment of the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various processes of the above-mentioned large model-based task processing method embodiment are implemented, and the same technical effect can be achieved. To avoid repetition, it will not be repeated here.

[0178] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0179] An embodiment of the present application also provides a computer program product, including a computer program, which implements the above-mentioned large model-based task processing method when executed by a processor.

[0180] The processor is the processor in the electronic device described in the above embodiment. The readable storage medium includes a computer readable storage medium, such as a computer read-only memory ROM, a random access memory RAM, a magnetic disk or an optical disk.

[0181] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0182] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0183] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

[0184] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0185] In the description of the present application, “plurality” means two or more.

[0186] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "illustrative embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0187] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.

Claims

1. A task processing method based on a large model, characterized in that: include: Get the voice information input by the user; Using a large model to perform intent recognition on the voice information to determine the task type; According to the task type, using the large model to control the virtual digital human to communicate with the user to determine task requirement information; Process the corresponding tasks according to the task requirement information and the large model.

2. The task processing method based on a large model according to claim 1, characterized in that: When the task type is a knowledge question answering type, the task requirement information includes question information, and the processing of the corresponding task according to the task requirement information and the large model includes: Retrieving knowledge matching the problem information in a preset knowledge database; generating answer information using the large model based on the retrieved knowledge and prompt engineering; The answer information is displayed to the user to conduct a knowledge question and answer session.

3. The task processing method based on a large model according to claim 1, characterized in that: When the task type is business handling, the task requirement information includes a filled business word slot, and according to the task type, the virtual digital human is controlled by the large model to communicate with the user to determine the task requirement information, including: Identify multiple business word slots; The large model is used to control the virtual digital person to have multiple rounds of dialogue with the user according to the business word slot, so as to guide the user to fill in the business word slot.

4. The task processing method based on a large model according to claim 3 is characterized in that: The using of the large model to control the virtual digital person to have multiple rounds of dialogue with the user according to the business word slot to guide the user to fill in the business word slot includes: Using the large model, determine the filling process of the current business word slot, the filling process includes user guidance instructions and the interface to be called; According to the filling process, controlling the virtual digital person to have multiple rounds of dialogue with the user to guide the user to fill the current business word slot; If the current business word slot is filled, the next business word slot is selected as the current business word slot, and the step of using the large model to determine the filling process of the current business word slot is returned to execution until all the business word slots are filled and the conversation is ended.

5. The task processing method based on a large model according to claim 3 is characterized in that: The processing of the corresponding task according to the task requirement information and the large model includes: The large model is used to handle corresponding services for the user according to the filled service word slots and the user database.

6. The large model-based task processing method according to any one of claims 1 to 5, characterized in that: The task processing method based on the large model also includes: Obtain industry knowledge sample sets, multi-task instruction sample sets and guided learning sample sets; Pre-training the large model using the industry knowledge sample set; Using the multi-task instruction sample set to perform fine-tuning training on the pre-trained large model; The guided learning sample set is used to perform reinforcement learning training on the large model after fine-tuning training.

7. A task processing device based on a large model, characterized in that: include: An acquisition module is used to acquire the voice information input by the user; A recognition module, used to use a large model to perform intent recognition on the voice information to determine the task type; A dialogue module, used to control the virtual digital human to dialogue with the user using the large model according to the task type, so as to determine task requirement information; An execution module is used to process corresponding tasks according to the task requirement information and the large model.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the large model-based task processing method according to any one of claims 1 to 6 is implemented.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the large model-based task processing method as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the large model-based task processing method as described in claims 1-6 is implemented.

Citation Information

Patent Citations

  • Information interaction system, method and device based on speech recognition

    CN107424601A

  • Intelligent interaction method, electronic device and storage medium

    CN108427722A

  • Intelligent robot integrating chat, knowledge and task questions and answers

    CN113515613A

  • Man-machine interaction method and device based on large language model and electronic equipment

    CN117787409A

  • Corpus data set construction method and device of academic large model, equipment and medium

    CN117851562A