Response methods and related devices

By receiving business requests, extracting input text, querying the intersection of the first and second tag sets, determining the third tag set, and querying response information according to priority, the problem of individual user differences and business diversity in existing response methods is solved, and personalized and accurate responses are achieved.

CN118656437BActive Publication Date: 2026-11-17MASHANG CONSUMER FINANCE CO LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202410537190.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-29
Publication Date
2026-11-17
Estimated Expiration
2044-04-29

AI Technical Summary

Technical Problem

Existing response methods ignore individual user differences and diverse business needs, resulting in response quality that fails to meet actual requirements.

Method used

By receiving business requests, extracting input text, querying the intersection of the pre-configured first tag set and second tag set, determining the third tag set, and querying response information according to priority, personalized responses can be achieved.

Benefits of technology

It improves the accuracy of responses, taking into account both the intent information reflected in the business requests of the target object and the differentiated needs of tags in the business, thus ensuring the personalization and accuracy of the responses.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118656437B_ABST
    Figure CN118656437B_ABST
Patent Text Reader

Abstract

This application provides a response method, apparatus, device, computer-readable storage medium, and computer program product. The method includes: receiving a service request from a target object; extracting input text from the service request; extracting intent information from the input text; querying a first tag set pre-configured for the intent information, the first tag set including tags pre-configured for the target service currently accessed by the target object; querying a second tag set of the target object, the second tag set including tags characterizing the features of the target object; determining the intersection of the second tag set and the first tag set as a third tag set; querying the priority of the tags in the third tag set; querying response information according to the priority; and responding to the service request according to the response information. This application improves the accuracy of the response.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a response method, apparatus, device, computer-readable storage medium, and computer program product. Background Technology

[0002] In response methods of related technologies, the intent identified from user input information is used to respond. However, this single response method ignores individual differences among users and the diverse needs of business, resulting in the quality of the response failing to meet actual requirements. Summary of the Invention

[0003] This application provides a response method, apparatus, device, computer-readable storage medium, and computer program product that can improve the accuracy of responses.

[0004] The technical solution of this application embodiment is implemented as follows:

[0005] This application provides a response method, the method comprising:

[0006] Receive a business request from the target object, and extract input text from the business request;

[0007] Intent information is extracted from the input text, and a first set of tags pre-configured for the intent information is queried. The first set of tags includes tags pre-configured for the target service currently used by the target object.

[0008] Query the second tag set of the target object, the second tag set including tags that characterize the features of the target object;

[0009] The intersection of the second tag set and the first tag set is determined and used as the third tag set;

[0010] Query the priority of the tags in the third tag set, and query the response information according to the priority;

[0011] The service request is responded to based on the response information.

[0012] This application provides a response device, including:

[0013] The request receiving module is used to receive business requests from the target object and extract input text from the business requests.

[0014] The request processing module is used to extract intent information from the input text and query a first tag set pre-configured for the intent information. The first tag set includes tags pre-configured for the target service currently used by the target object.

[0015] The request processing module is further configured to query a second tag set of the target object, the second tag set including tags that characterize the features of the target object;

[0016] The request processing module is further configured to determine the intersection of the second tag set and the first tag set as the third tag set;

[0017] The request processing module is also used to query the priority of the tags in the third tag set and query the response information according to the priority;

[0018] The request-response module is used to respond to the service request based on the response information.

[0019] This application provides an electronic device, the electronic device comprising:

[0020] Memory is used to store executable instructions for a computer;

[0021] The processor, when executing computer-executable instructions stored in the memory, implements the response method provided in the embodiments of this application.

[0022] This application provides a computer-readable storage medium storing a computer program or computer-executable instructions for implementing the response method provided in this application when executed by a processor.

[0023] This application provides a computer program product, including a computer program or computer executable instructions, which, when executed by a processor, implements the response method provided in this application.

[0024] The embodiments of this application have the following beneficial effects:

[0025] By obtaining intent information from the input text carried in the business request, and obtaining the intersection of the first set of tags pre-configured for the intent information and the second set of tags for the target object, a third set of tags is obtained. By calculating the intersection, the commonalities between the target object and the business are reflected, and personalized responses are achieved based on the differences in the tags of the target object. This takes into account both the intent information reflected in the business request of the target object and the differentiated needs of the target business in responding to different tags, thereby ensuring the accuracy of the response. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the architecture of the response system provided in the embodiments of this application;

[0027] Figure 2 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application;

[0028] Figure 3A This is a first flowchart illustrating the response method provided in an embodiment of this application;

[0029] Figure 3B This is a second flowchart illustrating the response method provided in the embodiments of this application;

[0030] Figure 3C This is a schematic diagram of the third process of the response method provided in the embodiments of this application;

[0031] Figure 3D This is a schematic diagram of the fourth process of the response method provided in the embodiments of this application;

[0032] Figure 3E This is a fifth flowchart illustrating the response method provided in the embodiments of this application;

[0033] Figure 3F This is a sixth flowchart illustrating the response method provided in the embodiments of this application;

[0034] Figure 3G This is a schematic diagram of the seventh process of the response method provided in the embodiments of this application;

[0035] Figure 3H This is the eighth flowchart of the response method provided in the embodiments of this application;

[0036] Figure 4A This is a first schematic diagram of the interface of the response system provided in the embodiments of this application;

[0037] Figure 4B This is a second schematic diagram of the interface of the response system provided in the embodiments of this application;

[0038] Figure 4C This is a third schematic diagram of the interface of the response system provided in the embodiments of this application;

[0039] Figure 4D This is a fourth schematic diagram of the interface of the response system provided in the embodiments of this application;

[0040] Figure 5 This is a schematic diagram of the overall flow of the response method provided in the embodiments of this application;

[0041] Figure 6 This is a schematic diagram of the interaction flow of the response method provided in the embodiments of this application.

[0042] It should be noted that the terms "first" and "second" mentioned above are only used to distinguish between different options and do not represent the degree of superiority or inferiority of the options or their priority in the implementation process. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0044] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0045] In the following description, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0046] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0047] Unless otherwise defined, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in the embodiments of this application is for the purpose of describing the embodiments of this application only and is not intended to limit this application.

[0048] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant national laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0049] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0050] 1) An answering system is a system that uses artificial intelligence technology for voice or text interaction and can automatically answer user questions. It understands the user's questions or needs and provides corresponding responses or solutions. Answering systems typically employ technologies such as natural language processing, machine learning, and deep learning, which continuously improve response accuracy and user experience through learning and optimization. Such systems have a wide range of applications, such as customer service robots, intelligent assistants, and smart homes.

[0051] For example, an answering system could be an Interactive Voice Response (IVR) system, an automated telephone system that uses pre-recorded or synthesized voice (also known as signpost voice or voice information) to interact with the user. IVR systems provide self-service over telephone lines, allowing users to navigate menus, query information, execute transactions, or receive notifications via voice commands or telephone keypad input.

[0052] 2) A slot refers to a location or field used to store specific information. Slots can be used to capture key information in the input text, such as entities, attributes, or constraints. In response systems, slots are typically used to store information provided by the target object so that the system can better understand the target object's intent and context and respond accordingly. For example, based on the target object's input text, multiple intent parameters are identified, and these parameters are combined to obtain intent information. The location where each intent parameter is placed is a slot.

[0053]

[0054] 3) Intent information, also known as semantics, refers to the intentions or purposes expressed by users when interacting with the response system. Intent information helps the system correctly respond to user needs and provide corresponding services. For example, in a response system, a user might say, "I want to book a high-speed rail ticket to city A tomorrow afternoon." This sentence conveys the user's intention to book a high-speed rail ticket and specifies the time and destination. The system can further refine the user's needs by recognizing and parsing the user's intent information, such as asking for specific train numbers, seat types, etc., in order to complete the order.

[0055] In dialogue systems, intent information can also be used in conjunction with slot information. Slot information specifies the user's specific requirements and goals, while intent information guides the system on how to handle these requirements and respond to the user. By effectively understanding and recognizing user intent information, the response system can answer user questions more accurately, improving interaction efficiency and user experience. In general, intent information acts as a bridge connecting user input and system response in natural language processing, helping the system accurately understand user needs and provide corresponding services. For example, combining the intent parameters corresponding to the four slots in the table above to form an intent is intent information, which can be represented as: Bill #Query #February #This Period.

[0056] 4) Voice information: In a voice-based response system, voice prompts are used to respond to the target. Voice information can be voice text or voice audio used to implement specific dialogue.

[0057] 5) Interaction nodes are a key component of the response system, responsible for handling specific interactions related to the target object's intent. Each interaction node is associated with a specific set of business logic to provide corresponding services or information when the relevant intent information is identified. For example, when multiple dialogues identify the target object's final intent, it is necessary to jump to the content of the interaction, that is, the interaction node is bound to the intent information and jumps to the corresponding business module (business line).

[0058] 6) The first tag set is a pre-configured set of tags corresponding to the intent information. One intent information corresponds to one first tag set. For example, if the target object provides feedback, the intent information can be represented as: Feedback#Complaint#Platform1#Member, and the corresponding first tag set can be represented as: [Feedback, Complaint, Suggestion].

[0059] 7) The second tag set, also known as the user tag set, is a description of the characteristics of the target object. It is usually used to classify and identify target objects so as to provide them with more accurate content and services.

[0060] In existing response methods, responses are based on intent derived from speech recognition. However, this approach ignores individual differences, leading to a one-dimensional response. The applicant discovered that in practical applications, for the same input text, the intent information corresponding to users with different tags can vary. In subsequent business processing, users with different tags may correspond to the same business process or different business processes. In other words, users with different tags expressing the same message (input text) may actually follow different processes and scenarios, resulting in a one-dimensional response problem in existing response methods.

[0061] To address the aforementioned issues, embodiments of this application provide a response method, apparatus, device, computer-readable storage medium, and computer program product that can improve the accuracy of responses.

[0062] The following describes exemplary applications of the devices provided in the embodiments of this application. The electronic devices provided in the embodiments of this application can be implemented as various types of terminal devices such as laptops, tablets, desktop computers, set-top boxes, smartphones, smart speakers, smartwatches, smart TVs, and vehicle terminals, or they can be implemented as servers.

[0063] See Figure 1 , Figure 1 This is a schematic diagram of the architecture of the response system provided in the embodiments of this application. Figure 1 The system involves server 100, terminal device 200-1, terminal device 200-2, and network 300. Terminal device 200-1 and terminal device 200-2 are connected to server 100 through network 300, which can be a wide area network, a local area network, or a combination of both.

[0064] In some embodiments, the embodiments of this application can be implemented collaboratively by a server and a terminal device. For example, terminal device 200-1 performs service configuration for a target service and sends the service configuration data to server 100. After receiving the service configuration data, server 100 returns the service configuration result to terminal device 200-1. Terminal device 200-2 sends a service request to server 100. After receiving the service request, server 100 obtains the corresponding response information through the response method provided in the embodiments of this application and sends the response information to terminal device 200-2, thereby responding to the service request.

[0065] In text-based human-computer interaction scenarios, the target object's business requests can carry text input, which can be entered through a browser or a dedicated client. This input can include the target object's inquiry content regarding the currently used business.

[0066] In voice-based human-computer interaction scenarios, the target object's business requests can carry voice audio, and the input text carried by the voice audio can be obtained through voice recognition.

[0067] In some embodiments, the terminal device or server can implement the response method provided in the embodiments of this application by running various computer-executable instructions or computer programs. For example, computer-executable instructions can be microprogram-level commands, machine instructions, or software instructions. Computer programs can be native programs or software modules in an operating system. In summary, the aforementioned computer-executable instructions can be any form of instruction, and the aforementioned computer programs can be any form of application program, module, or plug-in. Terminal devices include, but are not limited to, mobile phones, computers, smart voice interaction devices, smart home appliances, and in-vehicle terminals.

[0068] See Figure 2 , Figure 2 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Figure 2 The electronic device 400 shown may be the aforementioned server 100. Figure 2 The illustrated electronic device 400 includes at least one processor 410, a memory 430, and at least one network interface 420. The various components of the electronic device 400 are coupled together via a bus system 440. It is understood that the bus system 440 is used to implement communication between these components. In addition to a data bus, the bus system 440 also includes a power bus, a control bus, and a status signal bus. However, for clarity, ... Figure 2 The general labeled all buses as Bus System 440.

[0069] The processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0070] The memory 430 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 430 may optionally include one or more storage devices physically located away from the processor 410.

[0071] The memory 430 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), and the volatile memory may be random access memory (RAM). The memory 430 described in this application embodiment is intended to include any suitable type of memory.

[0072] In some embodiments, memory 430 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, as illustrated below.

[0073] Operating system 431 includes system programs for handling various basic system services and performing hardware-related tasks, such as the framework layer, core library layer, driver layer, etc., for implementing various basic business functions and handling hardware-based tasks;

[0074] The network communication module 432 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 including: Bluetooth, WiFi, and Universal Serial Bus (USB), etc.

[0075] In some embodiments, the apparatus provided in this application can be implemented in software. Figure 2 A response device 433 stored in memory 430 is shown. This device can be software in the form of programs and plug-ins, and includes the following software modules: a request receiving module 4331, a request processing module 4332, and a response module 4333. These modules are logically linked and can therefore be arbitrarily combined or further separated according to their implemented functions. The functions of each module will be described below.

[0076] The following will describe the response method provided in this application embodiment, using the server as the execution subject, with exemplary applications and implementations of the server provided in the embodiments of this application. See also Figure 3A , Figure 3A This is a first flowchart illustrating the response method provided in the embodiments of this application, which will be combined with... Figure 3A The steps shown are explained.

[0077] In step 101, a business request from the target object is received, and input text is extracted from the business request.

[0078] In some embodiments, receiving a service request from a target object is achieved by sending an incoming call to the target service's access number. The input text is extracted from the service request. For example, the user (target object) can input a service request in text form (corresponding to the input text) through an application, webpage, etc., or by dialing a specific number (access number) to send a service request (the service request includes the target object's voice / audio). This is typically done via an incoming phone call. After the user dials the specific number, the incoming request is received by the system. Once the service request (e.g., an incoming phone call) is received by the system, the system will receive the user's voice / audio. Next, the system needs to convert the voice / audio into text using speech recognition technology so that the system can understand the user's intent and needs. Speech recognition is an artificial intelligence technology that analyzes and processes audio signals to convert them into recognizable text content (input text).

[0079] In step 102, intent information is extracted from the input text, and a first set of tags pre-configured for the intent information is queried. The first set of tags includes tags pre-configured for the target service currently accessed by the target object.

[0080] Following the example above, when the target object sends a business request related to overdue business by dialing a specific number (access number), the first tag set pre-configured for the intent information (that is, the intent information related to overdue business) is queried. The first tag set can be represented as [overdue, member, overpayment, platform 1].

[0081] In some embodiments, see Figure 3B , Figure 3A The step 102 shown, which involves extracting intent information from the input text, can be achieved through steps 1021 to 1023, as explained below.

[0082] In step 1021, a first tag in the second tag set is determined. The first tag has a higher priority than the first set priority and corresponds to a specific intent type.

[0083] In some embodiments, the first tag can be queried from the intent configuration table of the database in the second tag set (the tag corresponding to the target object, such as the tag of membership, overdue, etc.). The first tag corresponds to a specific intent type. The priority of the first tag is higher than the first set priority. See Table 1. For example, if the priority of the first set priority is 3 out of 1 to 3, then the first tag can be any of the tags corresponding to priority 2 or priority 1 respectively (corresponding to "member" and "overdue" in Table 1). The first tag can also be the tag with the highest priority in the second tag set. For example, the first tag is "overdue". The specific intent type corresponding to the first tag refers to the intent type related to overdue business, such as "overdue inquiry" or "overdue processing".

[0084] For example, an intent configuration table used to characterize the relationship between the tags in the second tag set, the tag priorities, and the first intent recognition model can be represented as:

[0085] Table 1

[0086]

[0087]

[0088] For example, the tags (second tag set) corresponding to the target object can be obtained in the following ways: First, obtain historical data of the target object during the dialogue or related business, such as input text (including input text corresponding to the target object's voice audio), interaction methods, timestamps, business processing records, etc. This data can be obtained through log analysis or third-party data sources; Next, clean and preprocess the collected data, including removing noise, standardizing data formats, and removing sensitive personal information; Finally, use natural language processing technology to classify and tag the preprocessed data, such as obtaining tags corresponding to the target object through steps such as word segmentation, part-of-speech tagging, named entity recognition, and sentiment analysis, and store the target object and its corresponding tags in the database.

[0089] In step 1022, if a first intent recognition model corresponding to the first label is set, the first intent recognition model is used to recognize the input text to obtain intent parameters.

[0090] In some embodiments, a first intent recognition model is used to identify intent parameters belonging to a specific intent type (one or more types) from input text. The first intent recognition model corresponds to a specific intent type, and the number of first intent recognition models can be one or more. One first intent recognition model is used to identify intent parameters of a specific intent type from input text. The first intent recognition model is trained based on a first sample text, which represents intent information related to a first label with a priority higher than a first set priority. By training the first intent model in this way, it is specifically designed to identify intent parameters of a specific intent type. Compared with a general intent recognition model that can identify intent parameters of multiple intent types, it has higher recognition efficiency and accuracy.

[0091] For example, the first model for overdue payment tags collects the user's statements (corresponding to input text, including input text obtained through speech recognition of audio) and the corresponding true intent information, that is, semantic-intent correspondence. Furthermore, the same intent can correspond to multiple input texts. For example, if the intent is "deferred repayment," the corresponding multiple input texts could be such as "Can I defer repayment?", "How do I apply for deferred repayment?", "What is the process for deferred repayment?", etc.

[0092] Additionally, for example, when the second tag set contains the overdue tag, the target's expression of "delay" corresponds to the target's intention to apply for a deferred repayment. However, in scenarios without overdue payments (i.e., no overdue payments in the second tag set), the input text expressing the target's intention to apply for a deferred repayment will be identified from the input text, thus revealing the intent parameter of inquiring about deferred repayment. Therefore, different input texts are collected for different tags, and the model is trained to identify the input text to obtain intent parameters corresponding to different intent types, thereby obtaining different models (corresponding to Model 1, Model 2, and Model 3 in Table 1). Similarly, for the membership tag, if the second tag set contains the membership tag, the target's inquiry about membership-related services will be identified, thus revealing specific intent parameters related to membership services. In other words, a target with the membership tag in the second tag set can apply for membership benefits, while a target without the membership tag can only inquire about membership-related services.

[0093] In other embodiments, multiple tags can be combined (e.g., combining the "overdue" and "member" tags in Table 1 into a combined tag). For the combined tag, the user's speech can be combined (i.e., the input text corresponding to the user's speech) to train the corresponding model. Model 3 in Table 1 is a general speech. When the second tag set cannot be found or no first intent recognition model corresponding to the first tag is set, model 3 can be used as the second intent recognition model. Here, the specific implementation of recognizing the input text through the second intent recognition model to obtain the intent parameters can be found in the description of steps 1026 to 1028 below.

[0094] In some embodiments, the input text is identified by a first intent recognition model to obtain intent parameters. Taking the case where the first label in the above text is the highest priority label as an example, intent parameters related to the highest priority label and belonging to a specific intent type (1 or more types) are obtained.

[0095] For example, when the highest priority label in the second label set is "overdue", Model 1 (corresponding to the first intent recognition model) in Table 1 can be called to recognize the input text "Please help me check my February bill, thank you". The recognized intent parameters can be represented by Table 2:

[0096] Table 2

[0097] Slot Name main body action Parameter 1 Parameter 2 Slot value bill Query February -

[0098] Model 1 identifies and extracts slot values ​​(intent parameters) corresponding to slots (the positions where each intent parameter is placed) from the input text of the target object and fills them into the corresponding slots. When the input text of the target object is extracted based on the target object's speech audio, technologies such as speech recognition and natural language understanding will be involved. For example, firstly, the input text obtained from speech recognition is preprocessed, including word segmentation, stop word removal, and part-of-speech tagging. Next, named entity recognition technology can be used to identify specific entities (corresponding to intent parameters) in the input text, such as names, locations, and organizations. These entities usually correspond to predefined slots. Finally, based on the predefined slots (as shown in Table 2, four slots are predefined, including the name of each slot) and the entities identified in the input text, a certain rule or algorithm is used to determine which slot the entity should fill. These rules can be based on the context of the entity's appearance, syntactic structure, or specific keywords. For example, if the input text includes "Today's date is January 1st", then a rule can be predefined to match "January 1st" to the slot named "Date". In addition, the filling of some slots may depend on contextual information. For example, if the input text includes "I like to eat apples", the context can be used to determine whether "apple" should be filled into the "Fruit" slot or the "Food" slot.

[0099] In other embodiments, machine learning techniques such as decision trees, random forests, conditional random fields (CRFs), or neural networks can be used to train models on labeled datasets. These models can learn how to predict the most suitable slots based on the features of the input text, thereby obtaining the intent parameters corresponding to the input text. The embodiments of this application do not limit the specific method of obtaining intent parameters.

[0100] In step 1023, intent information is generated based on the identified intent parameters.

[0101] Following the example above, as shown in Table 2, the input text "Please help me check my bill for February, thank you" is recognized by the first intent recognition model. The resulting intent parameters include "bill", "query", and "February". The intent information can be represented as a combination of multiple intent parameters. For example, the intent information can be represented as: bill#query#February#*.

[0102] Through steps 1021 to 1023, a specific first intent recognition model is obtained based on the second tag set of the target object for targeted processing, thereby further obtaining intent parameters. Since the first intent recognition model is specifically designed to identify intent parameters corresponding to a specific intent type, it has higher recognition efficiency and accuracy compared to a general intent recognition model that can identify intent parameters corresponding to multiple intent types. This achieves the beneficial effect of improving the efficiency and accuracy of obtaining intent information of the target object.

[0103] In some embodiments, see Figure 3C , Figure 3B Step 1023 shown can be implemented through steps 10231 to 10234, which will be explained in detail below.

[0104] In step 10231, a pre-configured intent template is obtained, wherein the intent template includes multiple slots and the type of each slot.

[0105] In some embodiments, referring to Table 1, the intent template may include a preset number of slots and the name (type) of each slot. Different models may correspond to different templates (such as modifying the number of slots and the name of the slots).

[0106] In step 10232, the intent parameter is filled into the corresponding slot as the slot value.

[0107] In some embodiments, the intent parameters identified by the first intent recognition model are filled into the corresponding slots as slot values.

[0108] Following the example above, taking "overdue" as the highest priority label in the second label set and the first label as the first label, we can call Model 1 in Table 1 to perform intent recognition on the input text "Please help me check my bill for February, thank you". The obtained intent parameters can be represented by Table 2. For details on the implementation, please refer to the description of step 1022, which will not be repeated here.

[0109] In other embodiments, the intent parameters identified by the second intent recognition model are used as slot values ​​to fill the corresponding slots. The processing method here is similar to the slot value filling process of the first recognition model, so it will not be described again.

[0110] In step 10233, if there are no empty slots among the multiple slots, the filled intent template is used as the intent information.

[0111] Following the example above, when all four slots shown in Table 2 are filled with the corresponding intent parameters and there are no empty slots, the filled intent template will be used as the intent information.

[0112] In step 10234, if there are empty slots among the multiple slots, the slot value corresponding to the empty slot is queried based on the tags in the second tag set, and the slot value is filled into the empty slot. The filled intent template is used as intent information.

[0113] In some embodiments, after the intent parameter is filled into the corresponding slot as a slot value, if there is an empty slot among multiple slots, the slot value corresponding to the empty slot is queried from the slot-tag-slot filling value table (see Table 3) in the database based on the tags in the second tag set, and the slot value is filled into the empty slot. The filled intent template is then used as the intent information.

[0114] Continuing from the previous example, the intent parameters for the input text "Please help me check my February bills, thank you" include "bills," "check," and "February." Intent information can be represented as a combination of multiple intent parameters. Slot "Parameter 2" is empty. In this case, based on the tags in the second tag set, the slot value corresponding to the empty slot is queried, and the slot value is filled into the empty slot. The filled intent template is then used as the intent information. For example, the relationship between the tags in the second tag set, the slots, and the slot values ​​can be represented by Table 3:

[0115] Table 3

[0116] Slot Name Label Slot filling value main body Overdue Overdue Parameter 1 Platform 1 Platform 1 Member Parameter 2 Platform 1 Regular User Platform 1 Regular User Parameter 2 Platform 2 Member Platform 2 Member Parameter 2 Platform 2 Regular User Platform 2 Regular User main body member member Parameter 2 Overdue Repayment date

[0117] Assuming the tag in the second tag set is "overdue", by querying Table 3 to obtain the slot corresponding to the empty slot "Parameter 2" and filling it with the value "Repayment Date", the filled intent template can be represented as Table 4:

[0118] Table 4

[0119] Slot Name main body action Parameter 1 Parameter 2 Slot value bill Query February Repayment date

[0120] Intent information can be represented as a combination of multiple intent parameters: bill # query # February # repayment date. Here, Table 3 is only an example (single label correspondence). Table 3 can be adapted to specific business needs, such as adding a correspondence between slots and labels. For example, when the slot name is "subject", the corresponding labels may include "overdue" and "member". This application embodiment does not limit the specific correspondence between slot-label-slot filling value.

[0121] Through steps 10231 to 10234, when there is an empty slot, the slot value corresponding to the empty slot is queried based on the tags in the second tag set, and the slot value is filled into the empty slot. This achieves the beneficial effect of automatically improving the intent information of the target object when the intent parameters are complete, thereby improving the accuracy of the obtained intent information.

[0122] See also Figure 3A In step 103, a second tag set of the target object is queried. The second tag set includes tags that characterize the features of the target object.

[0123] In some embodiments, a second set of tags (including at least one second tag) for the target object can be queried from the object-tag table in the database based on the target object's ID. The second tag set includes multiple tags that characterize the features of the target object. For example, the object-tag table can be represented as Table 5:

[0124] Table 5

[0125] Target object ID Second tag set *****-1 Overdue; Platform 1 *****-2 Members; Platform 2 *****-3 Overdue; Member; Platform 1 *****-4 There are repayment records in the last 24 hours; Platform 1

[0126] In step 104, the intersection of the second tag set and the first tag set is determined and used as the third tag set.

[0127] For example, the first tag set can be represented as [overdue, member, overpayment], the second tag set can be represented as [overdue, platform 1], and the intersection of the second tag set and the first tag set is determined as the third tag set, which can be represented as [overdue].

[0128] In step 105, the priority of the tags in the third tag set is queried, and the response information is retrieved based on the priority.

[0129] In some embodiments, the priority of a tag in a third tag set can be queried from the intent configuration table (intent-tag-priority-response information) in the database, and the response information (voice information and interaction node) can be queried according to the priority. For example, the intent configuration table can be represented by Table 6:

[0130] Table 6

[0131]

[0132] Among them, the first semantic information "*#repayment#*#*" is the intention parameter including the repayment situation (corresponding to the slot value of "action" in Table 2 as "repayment"). Continuing from the example above, the third tag set is [overdue]. According to Table 6, the "overdue" tag is the highest priority tag in the third tag set, corresponding to voice information and business exit information. Here, the representation of the first semantic information is only an example. The embodiments of this application do not limit the representation of the first semantic information, nor the correspondence between the first semantic information and the tags (it can correspond to a single tag or multiple tags).

[0133] In some embodiments, see Figure 3D , Figure 3A Step 105 shown can be implemented through steps 1051 to 1052, which will be explained in detail below.

[0134] In step 1051, a second tag is determined in the third tag set, and the priority of the second tag is higher than the second set priority.

[0135] In some embodiments, the second label in the third label set is determined by the intent configuration table (Table 6) of the database. For example, the intent information “*#repayment#*#*” corresponds to the third label set including “overdue”, “payable this period” and “overpayment”. When the priority of the second setting is 3 out of 5, then the second label can be any one of the labels corresponding to priority 2 or priority 1 respectively (such as the “overdue” label corresponding to priority 1 and the “payable this period” label corresponding to priority 2 in Table 6).

[0136] In other embodiments, the second tag may also be the highest priority tag in the third tag set. The higher the priority of the tag, the more likely the response information corresponding to the tag can be applied in the target business. For example, when the third tag set corresponding to the intent information "*#repayment#*#*" includes "payable this period" and "overpayment", since the priority of "payable this period" is 2 and the priority of "overpayment" is 3, "payable this period" is the highest priority tag (i.e., the second tag) in the current third tag set.

[0137] In step 1052, the response information corresponding to the second tag is determined.

[0138] Following the example above, taking the case where the second tag is the highest priority tag in the third tag set as an example, we look up Table 6 and output the voice information and interaction node (whether it is a business exit and the name of the business exit) corresponding to the highest priority tag "This period is due" as the response information.

[0139] See also Figure 3A In step 106, the service request is responded to based on the response information.

[0140] In some embodiments, where the response information is in voice format, the response information includes voice information and interaction nodes. The voice information is used to express different types of dialogue, and the interaction nodes include service exit nodes and non-service exit nodes. See [link to documentation]. Figure 3E , Figure 3A Step 106 shown can be implemented through steps 1061 to 1062, which will be explained in detail below.

[0141] In step 1061, in response to the interaction node being a service exit node, voice information including transfer scripts is returned to the terminal device that sent the service request.

[0142] In some embodiments, when the interaction node is a service exit point, the terminal device sending the service request returns voice information including a transfer script and the name of the service exit node, as shown in Table 6. The voice information including the transfer script is such as "Okay, please explain the reason for the fee" or "Okay, please process this repayment." The name of the service exit node corresponding to the voice information can serve as a data transmission identifier to ensure that the voice information and the corresponding service request are correctly connected, thereby correctly transmitting data. Here, the voice information can be displayed on the interface of the terminal device sending the service request in the form of text corresponding to the voice, or it can be played by the terminal device sending the service request in the form of audio.

[0143] In step 1062, in response to the fact that the interaction node is not a service exit node, voice information including follow-up questions is returned to the terminal device that sent the service request.

[0144] In some embodiments, when the interaction node is not a service exit point, the terminal device that sends the service request returns voice information including follow-up questions, as shown in Table 6. For example, the voice information including follow-up questions could be "Do you want to inquire about the cause of the fee, or negotiate a repayment? Please tell me," or "Do you want to settle the payment early, check the repayment record, or check the repayment balance? Please state your question directly."

[0145] In some embodiments, see Figure 3F ,exist Figure 3A Before extracting intent information from the input text in step 102, the following steps 1024 to 1025 may also be performed, which are described in detail below.

[0146] In step 1024, the first sample text is used as input, and the intent parameters of the specific intent type related to the label in the first sample text are used as the expected output to train the initial intent recognition model, thereby obtaining the first intent recognition model corresponding to the label.

[0147] In some embodiments, the number of first intent recognition models can be one or more, and one first intent recognition model is used to recognize intent parameters of a specific intent type. For each tag, the following processes are performed: obtaining a first training text, wherein the first training text includes a first sample text and intent parameters of a specific intent type corresponding to the first sample; performing intent recognition processing on the first sample text using the initialized first intent recognition model to obtain the predicted intent parameters of the first sample text; obtaining a first loss value (e.g., obtaining the first loss value through a preset loss function (such as a cross-entropy loss function); updating the parameters of the initialized first intent recognition model using the first loss value to obtain the trained first intent recognition model.

[0148] For example, for the label "overdue", the corresponding first sample text is obtained, such as "Can I postpone repayment", "How to apply for postpone repayment", "What is the process for postpone repayment", etc. The intent parameters corresponding to the first sample text can be, for example, "Postponement #Repayment #*#*", "Postponement #Repayment #Application #*", "Postponement #Repayment #Process #*", etc. The initial intent recognition model, such as a recurrent neural network (RNN) or a long short-term memory network (LSTM), is trained. In this process, the model learns how to extract features from the input first sample text and map them to the corresponding intent parameters.

[0149] In step 1025, the second sample text is used as input, and the intent parameters of any intent type in the second sample text are used as the expected output to train the initialized intent recognition model, thereby obtaining the second intent recognition model.

[0150] In some embodiments, the second sample text may include multiple labels, each label corresponding to a specific intent type. A second training text is obtained, comprising the second sample text and the intent parameters corresponding to the second sample. The second sample text is processed by an initialized second intent recognition model to obtain predicted intent parameters. A second loss value is obtained using the intent parameters corresponding to the second sample and the predicted intent parameters of the second sample text (e.g., using a preset loss function, such as cross-entropy loss). The parameters of the initialized second intent recognition model are updated using the second loss value to obtain a trained second intent recognition model. This enables the second intent recognition model to recognize intent parameters for multiple specific intent types. Here, the multiple specific intent types include the specific intent types corresponding to each of the first intent recognition models mentioned above; that is, the specific intent types are a subset of the multiple specific intent types. In other words, the second intent recognition model can recognize the intent parameters for the specific intent types corresponding to all labels configured for the business.

[0151] In some embodiments, see Figure 3G In the absence of a first intent recognition model, intent information can be extracted from the input text through the following steps 1026 to 1028, which are explained in detail below.

[0152] In step 1026, a second intent recognition model is obtained. The second intent recognition model corresponds to multiple intent types, including a specific intent type.

[0153] In some embodiments, the second intent recognition model is used to obtain the intent parameter of any one of the multiple intent types (that is, the second intent recognition model can identify the intent parameters of the specific intent type corresponding to all the tags for the business configuration). The second intent recognition model is trained based on the second sample text. The intent type represented by the second sample text is more than that of the first sample text. When the second tag set of the target object is not obtained (including the case where the target object does not have a corresponding second tag set), or when the first intent recognition model corresponding to the first tag is not set, the second intent recognition model is obtained. The training of the second intent model can be referred to the description of step 1025 above, which will not be repeated here.

[0154] In step 1027, the second intent recognition model is used to recognize the input text to obtain intent parameters.

[0155] In some embodiments, the second intent recognition model is used to identify and extract the slot value (intent parameter) corresponding to the slot (the position where each intent parameter is placed) from the input text of the target object. For the specific implementation method, please refer to the description of step 1022 above, which will not be repeated here.

[0156] In step 1028, intent information is generated based on the identified intent parameters.

[0157] The intent parameters obtained by the second intent recognition model are filled into the corresponding slots. For the specific implementation method, please refer to the description of step 1022 above, which will not be repeated here.

[0158] Through steps 1026 to 1028, when the first intent recognition model corresponding to the label is not obtained, the corresponding intent information is obtained by using a general second intent recognition model. This achieves the beneficial effects of improving the comprehensiveness and applicability of the system, ensuring that the system can handle more diverse user requests, and enhancing the robustness and fault tolerance of the system. Even if a certain model has a problem or the data is incomplete, the system can still continue to provide services to the target object and improve user satisfaction.

[0159] In some embodiments, the first tag set is obtained from a third-party system, and the intent information, the first tag set, and the response information are pre-configured in a database. See [link to documentation]. Figure 3H ,exist Figure 3A Before step 101 shown, steps 107 to 109 can also be performed, which will be explained in detail below.

[0160] In step 107, a configuration page for the target service is sent to the terminal device.

[0161] In some embodiments, see Figures 4A to 4D It can be done Figures 4A to 4D The page shown allows for configuration of business functions, including... Figure 4A The interface shown is used to configure semantics (intent information), the first tag set corresponding to the intent information, and the response information corresponding to the tags in the first tag set (for example, when the response information is in voice form, the response information may include voice information number, voice information text, interaction node, and business line). Figure 4B Used to configure whether the interaction node is a business exit point, and the type of business exit node corresponding to the business exit point; Figure 4C Information used to configure voice information, such as voice information encoding, voice information text, and the business line to which the voice information belongs; Figure 4D This is used to release the IVR version corresponding to the target business that has been configured above, that is, to bind this version of the IVR system to the corresponding business line.

[0162] In step 108, service configuration data for the target service sent by the terminal device is received. The service configuration data includes intent information, a first tag set corresponding to the intent information, and response information corresponding to the tags in the first tag set.

[0163] In some embodiments, see Figure 4A , Figure 4A This is a first schematic diagram of the interface of the response system provided in the embodiments of this application, which can be accessed through... Figure 4A The interface shown allows configuration of semantics (intent information), the first tag set corresponding to the intent information, and the response information corresponding to the tags in the first tag set.

[0164] In step 109, the business configuration data is stored in the database.

[0165] In some embodiments, the business configuration data configured through the configuration page is stored in the database.

[0166] In some embodiments, where the second tag set does not intersect with the first tag set, the following implementation can also be performed: querying response information pre-configured for intent information; responding to service requests based on the pre-configured response information.

[0167] In some embodiments, see Figure 4A When the second tag set has no intersection with the first tag set, it is possible to obtain Figure 4A The semantic types shown are the voice information and interaction nodes corresponding to the general semantics, which are used as response information to the current business request.

[0168] Through steps 101 to 106, the intent information is obtained from the input text carried by the business request, and the intersection of the first tag set pre-configured for the intent information and the second tag set of the target object is obtained to obtain the third tag set. The response information corresponding to the second tag in the third tag set is then obtained to respond to the business request. This is equivalent to responding by combining the commonalities of the target object and the business. Compared with related technologies that respond solely based on intent information, this achieves personalized responses based on the differences in the tags of the target object, improving the flexibility of the response. On the other hand, by obtaining the response information based on the priority of the tags, the need for priority response to different tags in the business is taken into account, thus ensuring the accuracy of the response.

[0169] The response method provided in this application can be applied to various scenarios that require a response, including: (1) customer service: such as handling customer inquiries, such as bank balance inquiries, credit card bill inquiries, airline flight information inquiries, etc.; (2) entertainment and leisure: such as providing voice games, jokes or music, etc.; (3) self-service: such as providing 24 / 7 self-service, such as airline self-check-in, bank self-service, etc.; (4) navigation routes and planning: such as providing voice navigation and route planning services, etc.

[0170] The following example, using an interactive voice response system, illustrates an exemplary application of this application's embodiments in customer service scenarios, such as handling inquiries about bank card balances, loan applications, and overdue payment requests. (See also...) Figure 5 , Figure 5 This is a schematic diagram of the overall flow of the response method provided in this application embodiment. First, the speech recognition model is invoked (corresponding to steps 1021 and 1026 above). Next, the speech recognition model performs semantic recognition on the input text of the target object (see the description in step 1022 above) to obtain the semantic recognition result. Next, when there is an empty slot, the tag slot is filled (corresponding to the description in step 10234 above) to obtain a new semantic recognition result (corresponding to using the filled intent template as intent information in step 10234 above). Finally, the corresponding response information is obtained through tag result processing (corresponding to the descriptions in steps 104 to 105 above), thereby responding to the business request. See Figure 6 , Figure 6 This is a schematic diagram of the interaction flow of the response method provided in the embodiments of this application, which will be described in detail below.

[0171] As an example, the aforementioned interactive voice response system includes components such as an instruction component (IVR-Engine), a natural language processing component (IVR-NLP), a text-to-speech component (IVR-TTS), a development component (IVR-DM), an automatic speech recognition (ASR) component, a communication platform (Freeswitch, used to provide routing and interconnection communication for audio, text, or any other form of media, as well as to provide self-service IVR services), and a carrier gateway. Among these, the configuration of the response system by the developers can be pre-accepted through the IVR-DM (such as configuring semantics (intent information), the first tag set corresponding to the intent information, and the response information corresponding to the tags in the first tag set).

[0172] In step 201, the terminal device sends a service request, which carries the voice and audio of the target object.

[0173] In some embodiments, the target object initiates a business request (such as a bank balance inquiry, credit card bill inquiry, etc.) on a terminal device, for example, by sending a business request through an application, a webpage, or by dialing a specific number (access number). The terminal device can use a microphone or other device to collect the target object's voice audio and convert it into digital audio data. The terminal device transmits the business request (including voice audio data) to the server through a network, and the transmission method can be through the Internet, Wi-Fi, cellular network, etc.

[0174] In step 202, the server receives the service request and identifies the input text from the voice audio.

[0175] In some embodiments, the server receives a service request and converts the audio into text using speech recognition technology to obtain the input text.

[0176] As an example, you can use Freeswitch to call the ASR component, pass the real-time voice audio of the target object, and return the text corresponding to the audio (input text).

[0177] In step 203, the server extracts intent information from the input text and queries a first set of tags pre-configured for the intent information. The first set of tags includes tags pre-configured for the target service currently used by the target object.

[0178] In some embodiments, the specific implementation of extracting intent information from the input text and querying a first set of tags pre-configured for the intent information can be found in the description of step 102 above, and will not be repeated here.

[0179] As an example, IVR-Engine can call IVR-DM to extract intent information from the input text of the target object and the context of multiple interactions, thereby obtaining the semantics (intent information) of the target object and querying the first set of tags pre-configured for the intent information. Specifically, IVR-NLP accepts the call from IVR-DM, passes the input text of the target object, obtains the corresponding intent information and returns it to IVR-Engine.

[0180] In step 204, the server queries a second set of tags for the target object, which includes tags that characterize the features of the target object.

[0181] In some embodiments, the specific implementation of querying the second tag set of the target object can be found in the description of step 103 above, and will not be repeated here.

[0182] As an example, the IVR-Engine can be used to call an external system to query the second set of tags for a target object.

[0183] In step 205, the server determines the intersection of the second tag set and the first tag set to serve as the third tag set.

[0184] In some embodiments, the intersection of the second tag set and the first tag set is determined as a specific implementation of the third tag set. For details, please refer to the description of step 104 above, which will not be repeated here.

[0185] As an example, the intersection of the second tag set and the first tag set can be determined using IVR-DM.

[0186] In step 206, the server queries the priority of the tags in the third tag set and queries the response information based on the priority.

[0187] In some embodiments, the specific implementation of querying the priority of tags in the third tag set and querying the response information according to the priority can be found in the description of step 105 above, and will not be repeated here.

[0188] As an example, the priority of a tag can be determined through IVR-DM, and the content to be executed (such as voice information, interaction nodes, etc. included in the response information) can be determined based on the priority, and the content to be executed can be returned to IVR-Engine.

[0189] In step 207, the server sends a response message.

[0190] Here, you can call the IVR-Engine via Freeswitch to return the response information that needs to be interacted with the target object. For example, after the target object calls the number, return the audio of the voice information that needs to be played to interact with the target object. Specifically, the IVR-TTS accepts the call to the IVR-Engine, passes the text of the input voice information, and returns the audio of the corresponding voice information.

[0191] In step 208, the terminal device responds to the service request with response information.

[0192] In some embodiments, for example, if the input text corresponding to the business request is "Please help me check my bank card balance", the response information may include the voice message "Okay, check bank card balance", and the terminal device responds to the business request through the response information.

[0193] Steps 201 to 208 achieve the following: intent information is obtained from the input text carried in the business request; the intersection of a first tag set pre-configured for the intent information and a second tag set for the target object is obtained to obtain a third tag set; and the response information corresponding to the second tag in the third tag set is obtained to respond to the business request. This is equivalent to responding by combining the commonalities of the target object and the business. Compared with related technologies that respond solely based on intent information, this achieves personalized responses based on the differences in the target object's tags, improving the flexibility of the response. On the other hand, by obtaining response information based on the priority of tags, the need for prioritizing responses to different tags in the business is taken into account, thus ensuring the accuracy of the response.

[0194] The following description continues to illustrate the exemplary structure of the response device 433 provided in the embodiments of this application as a software module. In some embodiments, such as Figure 2 As shown, the software module stored in the response device 433 of the memory 430 may include:

[0195] The request receiving module 4331 is used to receive the business request of the target object and extract the input text from the business request.

[0196] The request processing module 4332 is used to extract intent information from the input text and query a first tag set pre-configured for the intent information. The first tag set includes tags pre-configured for the target service currently accessed by the target object.

[0197] In some embodiments, the request processing module 4332 is further configured to query a second tag set of the target object, the second tag set including tags that characterize the features of the target object.

[0198] In some embodiments, the request processing module 4332 is further configured to determine the intersection of the second tag set and the first tag set as the third tag set.

[0199] In some embodiments, the request processing module 4332 is further configured to query the priority of tags in the third tag set and query response information based on the priority.

[0200] The request response module 4333 is used to respond to the service request based on the response information.

[0201] In some embodiments, the request processing module 4332 is further configured to determine a first tag in the second tag set, wherein the first tag has a higher priority than a first set priority and the first tag corresponds to a specific intent type; if a first intent recognition model corresponding to the first tag is set, the first intent recognition model is used to recognize the input text to obtain intent parameters; and intent information is generated based on the recognized intent parameters.

[0202] In some embodiments, when the first intent recognition model is not set, the request processing module 4332 is further configured to obtain a second intent recognition model, the second intent recognition model corresponding to multiple intent types, the multiple intent types including the specific intent type; use the second intent recognition model to recognize the input text to obtain intent parameters; and generate intent information based on the recognized intent parameters.

[0203] In some embodiments, the request processing module 4332 is further configured to obtain a pre-configured intent template, wherein the intent template includes multiple slots and a type for each slot; fill the corresponding slot with the intent parameter as a slot value; if there are no empty slots among the multiple slots, use the filled intent template as intent information; if there are empty slots among the multiple slots, query the slot value corresponding to the empty slot based on the first tag in the second tag set, fill the empty slot with the slot value, and use the filled intent template as intent information.

[0204] In some embodiments, the request processing module 4332 is further configured to: acquire a first training text, wherein the first training text includes a first sample text and intent parameters of the specific intent type corresponding to the first sample; perform intent recognition processing on the first sample text using an initialized first intent recognition model to obtain predicted intent parameters of the first sample text; obtain a first loss value using the intent parameters corresponding to the first sample and the predicted intent parameters of the first sample text; update the parameters of the initialized first intent recognition model using the first loss value to obtain a trained first intent recognition model; acquire a second training text, wherein the second training text includes a second sample text and intent parameters corresponding to the second sample; perform intent recognition processing on the second sample text using an initialized second intent recognition model to obtain predicted intent parameters of the second sample text; obtain a second loss value using the intent parameters corresponding to the second sample and the predicted intent parameters of the second sample text; update the parameters of the initialized second intent recognition model using the second loss value to obtain a trained second intent recognition model.

[0205] In some embodiments, the request processing module 4332 is further configured to determine a second tag in the third tag set, wherein the priority of the second tag is higher than the second set priority; and determine the response information corresponding to the second tag.

[0206] In some embodiments, the response information includes voice information and interaction nodes. The voice information is information used to express different types of dialogue. The interaction nodes include service exit nodes and non-service exit nodes. The request response module 4333 is further configured to, in response to the interaction node being the service exit node, return voice information including transfer dialogue to the terminal device that sent the service request; and in response to the interaction node being the non-service exit node, return voice information including follow-up dialogue to the terminal device that sent the service request.

[0207] In some embodiments, when the second tag set and the first tag set do not intersect, the request processing module 4332 is further configured to query response information pre-configured for the intent information and respond to the service request according to the pre-configured response information.

[0208] In some embodiments, the intent information, the first tag set, and the response information are pre-configured in a database. The request processing module 4332 is further configured to send a configuration page for the target service to the terminal device; receive service configuration data for the target service sent by the terminal device, wherein the service configuration data includes the intent information, the first tag set corresponding to the intent information, and the response information corresponding to the tags in the first tag set; and store the service configuration data in the database.

[0209] This application provides a computer program product comprising a computer program or computer-executable instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer-executable instructions from the computer-readable storage medium and executes the computer-executable instructions, causing the electronic device to perform the response method described in this application.

[0210] This application provides a computer-readable storage medium storing computer-executable instructions or a computer program. When the computer-executable instructions or the computer program are executed by a processor, the processor will execute the response method provided in this application, for example, such as... Figure 3A The response method is shown.

[0211] In some embodiments, the computer-readable storage medium may be a memory such as RAM, ROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a variety of devices including one or any combination of the above-mentioned memories.

[0212] In some embodiments, computer-executable instructions may take the form of programs, software, software modules, scripts, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.

[0213] As an example, computer-executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., files that store one or more modules, subroutines, or code sections).

[0214] As an example, computer-executable instructions can be deployed to execute on a single electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed across multiple locations and interconnected via a communication network.

[0215] In summary, through the embodiments of this application, intent information is obtained from the input text carried by the business request, and the intersection of the first tag set pre-configured for the intent information and the second tag set of the target object is obtained to obtain the third tag set. By calculating the intersection, the commonalities between the target object and the business are reflected, and personalized responses are achieved based on the differences in the tags of the target object. This takes into account both the intent information reflected in the business request of the target object and the differentiated needs of the target business in responding to different tags, thereby ensuring the accuracy of the response.

[0216] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A method of responding, c h a r a c t e r i z e d b y, The method includes: Receive a business request from the target object, and extract input text from the business request; A first tag is determined from the second tag set of the target object, the first tag having a higher priority than a first set priority, and the first tag corresponding to a specific intent type; if a first intent recognition model corresponding to the first tag is set, the first intent recognition model is used to recognize the input text to obtain intent parameters; intent information is generated based on the recognized intent parameters. Query a first set of tags pre-configured for the intent information, the first set of tags including tags pre-configured for the target service currently accessed by the target object; Query the second tag set of the target object, the second tag set including tags that characterize the features of the target object; The intersection of the second tag set and the first tag set is determined and used as the third tag set; Query the priority of the tags in the third tag set, and query the response information according to the priority; The service request is responded to based on the response information.

2. The method of claim 1, wherein, In the absence of the first intent recognition model, the method further includes: Obtain a second intent recognition model, which corresponds to multiple intent types, including the specific intent type; The second intent recognition model is used to recognize the input text to obtain intent parameters; Intent information is generated based on the identified intent parameters.

3. The method of claim 1, wherein, The generation of intent information based on the identified intent parameters includes: Obtain a pre-configured intent template, wherein the intent template includes multiple slots and a type for each slot; Fill the corresponding slot with the intent parameter as the slot value; If there are no empty slots among the multiple slots, the filled intent template will be used as the intent information; If there is an empty slot among the multiple slots, the slot value corresponding to the empty slot is queried based on the tags in the second tag set, and the slot value is filled into the empty slot. The filled intent template is then used as intent information.

4. The method of claim 1, wherein, The querying of the priority of tags in the third tag set, and the querying of response information based on the priority, includes: Determine the second tag in the third tag set, where the priority of the second tag is higher than the second set priority; Determine the response information corresponding to the second tag.

5. The method of claim 1, wherein, The response information includes voice information and interaction nodes. The voice information is used to express different types of dialogue. The interaction nodes include business exit nodes and non-business exit nodes. The interaction nodes are used to process specific interactions related to the intent of the target object. Each interaction node is associated with a specific set of business logic. The step of responding to the service request based on the response information includes: In response to the fact that the interaction node is the service exit node, voice information including transfer script is returned to the terminal device that sent the service request; In response to the fact that the interaction node is the non-service exit node, the terminal device that sent the service request returns voice information including follow-up questions.

6. A response device, characterized by The device includes: The request receiving module is used to receive business requests from the target object and extract input text from the business requests. The request processing module is used to determine a first tag in the second tag set of the target object, wherein the first tag has a higher priority than a first set priority and the first tag corresponds to a specific intent type; if a first intent recognition model corresponding to the first tag is set, the first intent recognition model is used to recognize the input text to obtain intent parameters; intent information is generated based on the recognized intent parameters; and a first tag set pre-configured for the intent information is queried, wherein the first tag set includes tags pre-configured for the target service currently accessed by the target object. The request processing module is further configured to query a second tag set of the target object, the second tag set including tags that characterize the features of the target object; The request processing module is further configured to determine the intersection of the second tag set and the first tag set as the third tag set; The request processing module is also used to query the priority of the tags in the third tag set and query the response information according to the priority; The request-response module is used to respond to the service request based on the response information.

7. An electronic device, characterized in that, The electronic device includes: Memory is used to store executable instructions for a computer; A processor, when executing computer-executable instructions or computer programs stored in the memory, implements the response method according to any one of claims 1 to 5.

8. A computer-readable storage medium storing computer-executable instructions or a computer program, characterized in that, When the computer-executable instructions or computer program are executed by a processor, the response method according to any one of claims 1 to 5 is implemented.

9. A computer program product comprising computer-executable instructions or a computer program, characterized in that, When the computer-executable instructions or computer program are executed by a processor, the response method according to any one of claims 1 to 5 is implemented.

Citation Information

Patent Citations

  • Intelligent customer service response method, system and device and storage medium

    CN114817506A