Man-machine conversation process design method and system

Abstract

The invention discloses a man-machine conversation process design method. The method comprises the steps that a knowledge base manager configures a knowledge base; after knowledge base configuration is completed, a flow chart is built through a flow chart manager; arranging nodes of the flow chart in the flow chart manager through a drag-and-drop operation, and configuring logic among the nodes; after the flow chart is built, the verification tester executes a verification test on the built flow chart; and executing the large model auxiliary operation through the dialogue debugger. In addition, the invention also discloses a man-machine conversation process design system. By adopting the method and the system, conversation verbal skills, conversation logics and engines and matched models related to conversation interaction in the man-machine conversation flow can be flexibly configured, so that the development efficiency of a man-machine conversation process is greatly improved; the node dragging type interaction mode is added, so that the logic of the process design process is clearer, and the customization of the man-machine conversation process is more efficiently completed under the assistance of a large model.

Description

Technical Field

[0001] The present invention relates to the technical field, and in particular to a human-computer dialogue process design method and system. Background Art

[0002] In the prior art, the definition of the human-computer dialogue process is usually achieved by writing program code or filling in a form. Among them, the human-computer dialogue process is implemented by writing program code, which is inefficient, requires high programming skills of personnel, and is not convenient for industry training and application promotion; while the dialogue process is implemented by filling in a form, the flexibility of the dialogue logic is low, and the logical relationship of the dialogue cannot be intuitively displayed. Summary of the invention

[0003] Based on this, in order to solve the technical problems in the existing technology, a human-computer dialogue process design method is proposed, including:

[0004] The knowledge base manager configures the knowledge base; the knowledge base is located in the business platform;

[0005] After the knowledge base is configured, a flowchart is built through the flowchart manager. The flowchart consists of one or more nodes and the logical configuration of the nodes. In the flowchart manager, the nodes of the flowchart are arranged and the logic between the nodes is configured by dragging and dropping.

[0006] After the flowchart is built, the verification tester performs a verification test on the built flowchart.

[0007] In one embodiment, the knowledge base stores the conversation knowledge background information of the voice robot, which is the knowledge basis for ensuring that the voice robot recognizes and understands customer conversations. In the subsequent execution of the human-computer dialogue process, the voice robot determines the node jump direction based on the conversation knowledge background information hit in the knowledge base.

[0008] In one embodiment, the verification tester verifies the syntax and configuration correctness of the flowchart, and performs a comprehension test after confirming that the syntax and configuration of the flowchart are correct;

[0009] In the comprehension test, the verification tester simulates a customer call to test the voice robot's intent recognition performance in the human-computer dialogue process. When problems are found in customer intent recognition, the verification tester feeds the problems back to the flowchart manager. The flowchart manager readjusts and builds the flowchart based on the feedback content, thereby optimizing the flowchart.

[0010] In one embodiment, in the human-computer dialogue process, after the voice robot enters a node in the flowchart, the recognition of the customer's intention includes one or more matching operations performed in sequence, including node branch intention judgment operation, common question answering intention judgment operation, exception handling operation, and fallback operation;

[0011] Among them, when executing the node branch intention judgment operation, each node is configured with one or more branch routes, each branch route corresponds to a customer intention, when the voice robot recognizes the intention in the branch during the conversation, it means that the intention is hit, the voice robot executes the operation in the node, and jumps out of the current node according to the branch corresponding to the intention, otherwise it enters the common question answering intention judgment operation;

[0012] Among them, when executing the FAQ intention judgment operation, each node is configured with one or more FAQs. When the corresponding intent is not matched in the node branch, the intent configuration in each FAQ is checked in turn. When a match is found, the FAQ content is broadcasted, otherwise the exception handling operation is entered;

[0013] Among them, performing abnormal operations includes:

[0014] Return to the specified node and re-execute the process; or hang up, that is, the voice robot announces the hang-up words and then hangs up the call; or transfer to manual, that is, the voice robot transfers the current call to a manual seat for processing;

[0015] Among them, anomalies include no matching intention, the customer not speaking, abnormal customer emotions, and abnormal noise;

[0016] Among them, when the node branch intention judgment operation, the common question answering intention judgment operation, and the exception handling fail to solve the customer needs, the fallback operation is triggered.

[0017] In one embodiment, the dialogue debugger configures the big model according to the business background of the current human-computer dialogue process; the dialogue debugger inputs the business description and configuration requirements in the business platform into the big model for configuration;

[0018] The dialogue debugger calls the configured big model to perform big model auxiliary generation operations in the current business context to assist in generating the required content;

[0019] The dialogue debugger transmits the assisted generated content back to the business platform and stores it in the knowledge base.

[0020] In addition, in order to solve the technical problems in the prior art, a human-computer dialogue process design system is proposed, including a knowledge base, a knowledge base manager, a flow chart manager, and a verification tester;

[0021] Wherein, the knowledge base manager is used to configure the knowledge base; the knowledge base is located in the business platform;

[0022] The flowchart manager is used to build a flowchart after the knowledge base configuration is completed. The flowchart consists of one or more nodes and the logical configuration of the nodes. In the flowchart manager, the nodes of the flowchart are arranged and the logic between the nodes is configured by dragging and dropping.

[0023] Among them, the verification tester is used to perform verification tests on the constructed flowchart after the flowchart is completed.

[0024] In one embodiment, the knowledge base includes an intention base, an interruption word base, a synonym word base, a hot word base, a frequently asked questions answer base, and an exception handling base;

[0025] The intent library contains one or more customer intents that the current voice robot needs to understand. The intent is the key meaning of a sentence spoken by the customer. The voice robot associates the customer's speech content with the fixed intent through keyword recognition and regular extraction, thereby identifying and understanding the customer's speech content;

[0026] The interruption word library includes one or more keywords of the customer's intention to stop the voice robot from continuing to speak. After the voice robot recognizes such keywords, it stops talking to the customer, thereby obtaining a result that the customer interrupts the voice robot in the human-machine dialogue.

[0027] The synonym database contains one or more corresponding combinations of standard words and synonyms, where one standard word corresponds to one or more synonyms. The voice robot uses the synonyms in the synonym database to deal with the situation where the same intention has multiple expressions in human-computer dialogue.

[0028] The hot word library contains one or more proper nouns, which are specific proper nouns in a specific business. The voice robot uses the hot words in the hot word library to deal with complex and specific proper nouns in human-computer dialogue.

[0029] The FAQ database contains one or more FAQs, which are questions frequently asked by customers in the current business and the corresponding voice robot answers. When the voice robot recognizes the customer's intention and matches the answer in the FAQ database, the voice robot performs corresponding dialogue operations according to the configuration.

[0030] Among them, the exception handling library contains one or more intentions that the voice robot has not recognized. The exception refers to the situation in which the voice robot fails to recognize the intention stored in the knowledge base during the human-computer dialogue, resulting in the inability to continue the process; the exception handling library also contains one or more operations for handling exceptions to ensure that the voice robot fully executes the human-computer dialogue process.

[0031] In one embodiment, the nodes of the flowchart include a start node, a voice interaction node, a transfer to manual node, a text message node, a sub-process node, a branch node, a transfer to external line node, and a script node;

[0032] Among them, the start node is the start node of the human-computer dialogue process. When the start node is executed, the human-computer dialogue process starts;

[0033] When executing to the voice interaction node, the voice robot broadcasts different words according to the customer's intention;

[0034] When the call is transferred to a human agent, the voice robot transfers the call to the configured agent group, and the corresponding human agent in the agent group takes over the call.

[0035] Among them, when executing to the SMS node, the voice robot sends a text message to the customer;

[0036] When the data interaction node is executed, the voice robot triggers the interface call and sends a data query request to the outside, and replies the queried data information to the customer;

[0037] Among them, when the needs raised by the customer cannot be solved by a short answer but need to be solved by a short process, the sub-process node is executed to solve the customer's needs through the sub-process;

[0038] Among them, the branch node is a judgment node. When executing to the branch node, the voice robot determines which branch to execute according to the intention;

[0039] Among them, when executing to the external line node, the voice robot transfers the call to the external phone;

[0040] Among them, when executing to the script node, the voice robot executes the configured script code;

[0041] Among them, when the end node is executed, the human-computer dialogue process ends.

[0042] In one embodiment, the human-computer dialogue process design system further includes a dialogue debugger, which configures the big model according to the business background of the current human-computer dialogue process; the dialogue debugger inputs the business description and configuration requirements in the business platform into the big model for configuration;

[0043] The dialogue debugger calls the configured big model to perform big model auxiliary generation operations in the current business context to assist in generating the required content;

[0044] The dialogue debugger transmits the assisted generated content back to the business platform and stores it in the knowledge base.

[0045] In one embodiment, the large model auxiliary operations performed by the dialog debugger specifically include:

[0046] The dialogue debugger assists in generating the speech in the voice interaction node. That is, the dialogue debugger generates one or more texts of the speech broadcast by the voice robot based on the current business background and node name using the big model, and stores them in the knowledge base.

[0047] The dialogue debugger assists in generating keywords in the intent library. That is, the dialogue debugger generates one or more keywords that express specific intents of customers based on the current business context and intent name using the big model, and stores them in the intent library of the knowledge base.

[0048] The dialogue debugger assists in generating similar questions in the intent library. That is, based on the current business context and intent name, the dialogue debugger uses the big model to generate one or more interrogative sentences that express specific intents of the customer, and stores them in the knowledge base.

[0049] The dialogue debugger assists in generating common questions and their answers. That is, the dialogue debugger uses a large model to generate one or more questions frequently asked by customers in the current business and their corresponding answers based on the current business context, and stores them in the common questions and answers library of the knowledge base; one common question corresponds to one or more answers.

[0050] Implementing the embodiments of the present invention will have the following beneficial effects:

[0051] The present invention adds a node dragging interaction mode to the traditional human-computer dialogue process design method to make the logic of the entire process design process clearer, and enables users to complete the customization of the human-computer dialogue process more efficiently with the assistance of a large model. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] in:

[0054] Figure 1 A schematic diagram of a human-computer dialogue process design method in the present invention;

[0055] Figure 2 A schematic diagram of a human-computer dialogue process design system in the present invention;

[0056] Figure 3 This is a schematic diagram of the customer intention recognition process in the present invention;

[0057] Figure 4 A schematic diagram of the dialog debugger of the present invention performing large model auxiliary operations. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] like Figure 1 As shown, the present invention discloses a method for designing a human-computer dialogue process, comprising:

[0060] The knowledge base manager configures the knowledge base; the knowledge base is located in the business platform;

[0061] The knowledge base stores the conversation knowledge background information of the voice robot, which is the knowledge basis for ensuring that the voice robot can recognize and understand the customer conversation.

[0062] In the subsequent execution of the human-computer dialogue process, the voice robot determines the node jump direction based on the dialogue knowledge background information it hits in the knowledge base;

[0063] Configuring the knowledge base is the first step in designing the entire human-computer dialogue process;

[0064] In particular, the knowledge base includes an intention base, an interruption base, a synonym base, a hot word base, a frequently asked questions base, and an exception handling base;

[0065] The intent library contains one or more customer intents that the current voice robot needs to understand. The intent is the key meaning of a sentence spoken by the customer. The voice robot associates the customer's speech content with the fixed intent through keyword recognition and regular extraction, thereby identifying and understanding the customer's speech content;

[0066] The interruption word library includes one or more keywords of the customer's intention to stop the voice robot from continuing to speak. After the voice robot recognizes such keywords, it stops talking to the customer, thereby obtaining a result that the customer interrupts the voice robot in the human-machine dialogue.

[0067] The synonym database contains one or more corresponding combinations of standard words and synonyms, where one standard word corresponds to one or more synonyms. The voice robot uses the synonyms in the synonym database to deal with the situation where the same intention has multiple expressions in human-computer dialogue.

[0068] The hot word library contains one or more proper nouns, which are specific proper nouns in a specific business. The voice robot uses the hot words in the hot word library to deal with complex and specific proper nouns in human-computer dialogue.

[0069] Among them, the Frequently Asked Questions (FAQ) library contains one or more FAQs, which are questions frequently asked by customers in the current business and the corresponding voice robot answers; when the voice robot recognizes the customer's intention and matches the answer in the FAQ library, the voice robot performs corresponding dialogue operations according to the configuration; for example, the dialogue operation includes the voice robot broadcasting a section of answering words to the customer's question according to the configured FAQ;

[0070] Among them, the exception handling library contains one or more intents that the voice robot has not recognized. The exception refers to the situation in which the voice robot fails to recognize the intent stored in the knowledge base during the human-computer dialogue, resulting in the inability to continue the process; the exception handling library also contains one or more operations for handling exceptions, which are used to ensure that the voice robot fully executes the human-computer dialogue process;

[0071] After the knowledge base is configured, a flowchart is built through the flowchart manager. The flowchart consists of one or more nodes and the logical configuration of the nodes.

[0072] Among them, in the flowchart manager, the nodes of the flowchart are arranged and the logic between the nodes is configured by dragging and dropping;

[0073] In particular, in complex business scenarios, the nodes in the flowchart are arranged and logically configured through code configuration to complete the definition of the voice robot;

[0074] In particular, the nodes of the flowchart include a start node, a voice interaction node, a transfer to manual node, a text message node, a sub-process node, a branch node, a transfer to external line node, and a script node;

[0075] Among them, the start node is the start node of the human-computer dialogue process. When the start node is executed, the human-computer dialogue process starts;

[0076] When executing to the voice interaction node, the voice robot broadcasts different words according to the customer's intention;

[0077] When the call is transferred to a human agent, the voice robot transfers the call to the configured agent group, and the corresponding human agent in the agent group takes over the call.

[0078] Among them, when executing to the SMS node, the voice robot sends a text message to the customer;

[0079] When the data interaction node is executed, the voice robot triggers the interface call and sends a data query request to the outside, and replies the queried data information to the customer;

[0080] Among them, when the needs raised by the customer cannot be solved by a short answer but need to be solved by a short process, the sub-process node is executed to solve the customer's needs through the sub-process;

[0081] Among them, the branch node is a judgment node. When executing to the branch node, the voice robot determines which branch to execute according to the intention;

[0082] Among them, when executing to the external line node, the voice robot transfers the call to the external phone;

[0083] Among them, when executing to the script node, the voice robot executes the configured script code;

[0084] For example, a calculator is written using script code, and the voice robot executes the script code to calculate business-related values ​​and broadcasts them to the user;

[0085] Among them, when the end node is executed, the human-computer dialogue process ends;

[0086] After the flowchart is built, the verification tester performs verification tests on the built flowchart;

[0087] In particular, the verification tester verifies the syntax and configuration correctness of the flowchart, and performs a comprehension test after confirming that the syntax and configuration of the flowchart are correct;

[0088] In the comprehension test, the verification tester simulates customer calls to test the voice robot's intention recognition performance in the human-computer dialogue process. When problems are found in customer intention recognition, the verification tester feeds the problems back to the flowchart manager, which readjusts and builds the flowchart based on the feedback content, thereby optimizing the flowchart.

[0089] In particular, if Figure 3 As shown, in the human-computer dialogue process, after the voice robot enters the node of the flowchart, the recognition of the customer's intention includes one or more matching operations performed in sequence, including node branch intention judgment operation, common question answering intention judgment operation, exception handling operation, and fallback operation;

[0090] Among them, when executing the node branch intention judgment operation, each node is configured with one or more branch routes, each branch route corresponds to a customer intention, when the voice robot recognizes the intention in the branch during the conversation, it means that the intention is hit, the voice robot executes the operation in the node, and jumps out of the current node according to the branch corresponding to the intention, otherwise it enters the common question answering intention judgment operation;

[0091] Among them, when executing the FAQ intention judgment operation, each node is configured with one or more FAQs. When the corresponding intent is not matched in the node branch, the intent configuration in each FAQ is checked in turn. When a match is found, the FAQ content is broadcasted, otherwise the exception handling operation is entered;

[0092] The exception handling operations include:

[0093] Return to the specified node and re-execute the process; or hang up, that is, the voice robot announces the hang-up words and then hangs up the call; or transfer to manual, that is, the voice robot transfers the current call to a manual seat for processing;

[0094] In particular, anomalies include no matching intention, the customer not speaking, abnormal customer emotions, and abnormal noise;

[0095] Exception handling operations can be manually configured as needed;

[0096] Among them, when the node branch intention judgment operation, the common question answering intention judgment operation, and the exception handling fail to solve the customer's needs, the fallback operation is triggered;

[0097] Usually, the bottom-line operation is an on-hook operation;

[0098] In particular, in the human-computer dialogue process design method, large model auxiliary operations are performed through the dialogue debugger, such as Figure 4 As shown, specifically including:

[0099] The dialogue debugger configures the large model according to the business context of the current human-computer dialogue process;

[0100] Specifically, the dialogue debugger inputs the business description and configuration requirements in the business platform into the large model for configuration;

[0101] In view of this, since the knowledge background of the big model is open, before using the big model, it is necessary to configure the relevant business background of the voice robot (equivalent to Prompt);

[0102] The dialogue debugger calls the configured big model to perform big model auxiliary generation operations in the current business context to assist in generating the required content;

[0103] The dialogue debugger transmits the assisted generated content back to the business platform and stores it in the knowledge base;

[0104] In particular, the large model auxiliary operations performed by the dialogue debugger include:

[0105] The dialogue debugger assists in generating the speech in the voice interaction node. That is, the dialogue debugger generates one or more texts of the speech broadcast by the voice robot based on the current business background and node name using the big model, and stores them in the knowledge base.

[0106] The dialogue debugger assists in generating keywords in the intent library. That is, the dialogue debugger generates one or more keywords that express specific intents of customers based on the current business context and intent name using the big model, and stores them in the intent library of the knowledge base.

[0107] Enriching the intent library can improve the accuracy of the voice robot's intent recognition;

[0108] The dialogue debugger assists in generating similar questions in the intent library. That is, based on the current business context and intent name, the dialogue debugger uses the big model to generate one or more interrogative sentences that express specific intents of the customer, and stores them in the knowledge base.

[0109] In particular, the voice robot uses similar questions and keywords in the intent library to identify customer intent. Since the customer's speech content is uncontrollable and keywords are scattered throughout a sentence, the voice robot uses similar questions to judge and identify customer intent from the dimension of the entire sentence.

[0110] The dialogue debugger assists in generating common questions and their answers. That is, the dialogue debugger uses a large model to generate one or more questions frequently asked by customers in the current business and their corresponding answers based on the current business context, and stores them in the common questions and answers library of the knowledge base; one common question corresponds to one or more answers.

[0111] In addition, if Figure 2 As shown, the present invention also discloses a human-computer dialogue process design system, including a knowledge base, a knowledge base manager, a flow chart manager, and a verification tester;

[0112] Wherein, the knowledge base manager is used to configure the knowledge base; the knowledge base is located in the business platform;

[0113] The knowledge base stores the conversation knowledge background information of the voice robot, which is the knowledge basis for ensuring that the voice robot can recognize and understand the customer conversation.

[0114] In the subsequent execution of the human-computer dialogue process, the voice robot determines the node jump direction based on the dialogue knowledge background information it hits in the knowledge base;

[0115] Configuring the knowledge base is the first step in designing the entire human-computer dialogue process;

[0116] In particular, the knowledge base includes an intention base, an interruption base, a synonym base, a hot word base, a frequently asked questions base, and an exception handling base;

[0117] The intent library contains one or more customer intents that the current voice robot needs to understand. The intent is the key meaning of a sentence spoken by the customer. The voice robot associates the customer's speech content with the fixed intent through keyword recognition and regular extraction, thereby identifying and understanding the customer's speech content;

[0118] The interruption word library includes one or more keywords of the customer's intention to stop the voice robot from continuing to speak. After the voice robot recognizes such keywords, it stops talking to the customer, thereby obtaining a result that the customer interrupts the voice robot in the human-machine dialogue.

[0119] The synonym database contains one or more corresponding combinations of standard words and synonyms, where one standard word corresponds to one or more synonyms. The voice robot uses the synonyms in the synonym database to deal with the situation where the same intention has multiple expressions in human-computer dialogue.

[0120] The hot word library contains one or more proper nouns, which are specific proper nouns in a specific business. The voice robot uses the hot words in the hot word library to deal with complex and specific proper nouns in human-computer dialogue.

[0121] Among them, the Frequently Asked Questions (FAQ) library contains one or more FAQs, which are questions frequently asked by customers in the current business and the corresponding voice robot answers; when the voice robot recognizes the customer's intention and matches the answer in the FAQ library, the voice robot performs corresponding dialogue operations according to the configuration; for example, the dialogue operation includes the voice robot broadcasting a section of answering words to the customer's question according to the configured FAQ;

[0122] Among them, the exception handling library contains one or more intents that the voice robot has not recognized. The exception refers to the situation in which the voice robot fails to recognize the intent stored in the knowledge base during the human-computer dialogue, resulting in the inability to continue the process; the exception handling library also contains one or more operations for handling exceptions, which are used to ensure that the voice robot fully executes the human-computer dialogue process;

[0123] The flowchart manager is used to build a flowchart after the knowledge base configuration is completed. The flowchart consists of one or more nodes and the logical configuration of the nodes.

[0124] In the flowchart manager, the nodes of the flowchart are arranged and the logic between the nodes is configured by dragging and dropping;

[0125] In particular, in complex business scenarios, the nodes in the flowchart are arranged and logically configured through code configuration to complete the definition of the voice robot;

[0126] In particular, the nodes of the flowchart include a start node, a voice interaction node, a transfer to manual node, a text message node, a sub-process node, a branch node, a transfer to external line node, and a script node;

[0127] Among them, the start node is the start node of the human-computer dialogue process. When the start node is executed, the human-computer dialogue process starts;

[0128] When executing to the voice interaction node, the voice robot broadcasts different words according to the customer's intention;

[0129] When the call is transferred to a human agent, the voice robot transfers the call to the configured agent group, and the corresponding human agent in the agent group takes over the call.

[0130] Among them, when executing to the SMS node, the voice robot sends a text message to the customer;

[0131] When the data interaction node is executed, the voice robot triggers the interface call and sends a data query request to the outside, and replies the queried data information to the customer;

[0132] Among them, when the needs raised by the customer cannot be solved by a short answer but need to be solved by a short process, the sub-process node is executed to solve the customer's needs through the sub-process;

[0133] Among them, the branch node is a judgment node. When executing to the branch node, the voice robot determines which branch to execute according to the intention;

[0134] Among them, when executing to the external line node, the voice robot transfers the call to the external phone;

[0135] Among them, when executing to the script node, the voice robot executes the configured script code;

[0136] For example, a calculator is written using script code, and the voice robot executes the script code to calculate business-related values ​​and broadcasts them to the user;

[0137] Among them, when the end node is executed, the human-computer dialogue process ends;

[0138] Among them, the verification tester is used to perform verification tests on the constructed flowchart after the flowchart is constructed;

[0139] In particular, the verification tester verifies the syntax and configuration correctness of the flowchart, and performs a comprehension test after confirming that the syntax and configuration of the flowchart are correct;

[0140] In the comprehension test, the verification tester simulates customer calls to test the voice robot's intention recognition performance in the human-computer dialogue process. When problems are found in customer intention recognition, the verification tester feeds the problems back to the flowchart manager, which readjusts and builds the flowchart based on the feedback content, thereby optimizing the flowchart.

[0141] In particular, if Figure 3 As shown, in the human-computer dialogue process, after the voice robot enters the node of the flowchart, the recognition of the customer's intention includes one or more matching operations performed in sequence, including node branch intention judgment operation, common question answering intention judgment operation, exception handling operation, and fallback operation;

[0142] Among them, when executing the node branch intention judgment operation, each node is configured with one or more branch routes, each branch route corresponds to a customer intention, when the voice robot recognizes the intention in the branch during the conversation, it means that the intention is hit, the voice robot executes the operation in the node, and jumps out of the current node according to the branch corresponding to the intention, otherwise it enters the common question answering intention judgment operation;

[0143] Among them, when executing the FAQ intention judgment operation, each node is configured with one or more FAQs. When the corresponding intent is not matched in the node branch, the intent configuration in each FAQ is checked in turn. When a match is found, the FAQ content is broadcasted, otherwise the exception handling operation is entered;

[0144] The exception handling operations include:

[0145] Return to the specified node and re-execute the process; or hang up, that is, the voice robot announces the hang-up words and then hangs up the call; or transfer to manual, that is, the voice robot transfers the current call to a manual seat for processing;

[0146] In particular, anomalies include no matching intention, the customer not speaking, abnormal customer emotions, and abnormal noise;

[0147] Exception handling operations can be manually configured as needed;

[0148] Among them, when the node branch intention judgment operation, the common question answering intention judgment operation, and the exception handling fail to solve the customer's needs, the fallback operation is triggered;

[0149] Usually, the bottom-line operation is an on-hook operation;

[0150] In particular, in the human-computer dialogue process design method, large model auxiliary operations are performed through the dialogue debugger, such as Figure 4 As shown, specifically including:

[0151] The human-computer dialogue process design system also includes a dialogue debugger, which configures a large model according to the business background of the current human-computer dialogue process;

[0152] Specifically, the dialogue debugger inputs the business description and configuration requirements in the business platform into the large model for configuration;

[0153] In view of this, since the knowledge background of the big model is open, before using the big model, it is necessary to configure the relevant business background of the voice robot (equivalent to Prompt);

[0154] The dialogue debugger calls the configured big model to perform big model auxiliary generation operations in the current business context to assist in generating the required content;

[0155] The dialogue debugger transmits the assisted generated content back to the business platform and stores it in the knowledge base;

[0156] In particular, the large model auxiliary operations performed by the dialogue debugger include:

[0157] The dialogue debugger assists in generating the speech in the voice interaction node. That is, the dialogue debugger generates one or more texts of the speech broadcast by the voice robot based on the current business background and node name using the big model, and stores them in the knowledge base.

[0158] The dialogue debugger assists in generating keywords in the intent library. That is, the dialogue debugger generates one or more keywords that express specific intents of customers based on the current business context and intent name using the big model, and stores them in the intent library of the knowledge base.

[0159] Enriching the intent library can improve the accuracy of the voice robot's intent recognition;

[0160] The dialogue debugger assists in generating similar questions in the intent library. That is, based on the current business context and intent name, the dialogue debugger uses the big model to generate one or more interrogative sentences that express specific intents of the customer, and stores them in the knowledge base.

[0161] In particular, the voice robot uses similar questions and keywords in the intent library to identify customer intent. Since the customer's speech content is uncontrollable and keywords are scattered throughout a sentence, the voice robot uses similar questions to judge and identify customer intent from the dimension of the entire sentence.

[0162] The dialogue debugger assists in generating common questions and their answers. That is, the dialogue debugger uses a large model to generate one or more questions frequently asked by customers in the current business and their corresponding answers based on the current business context, and stores them in the common questions and answers library of the knowledge base; one common question corresponds to one or more answers.

[0163] The present invention adopts a human-computer dialogue process design method and system that mainly uses drag-and-drop flowchart construction, supplemented by code configuration, and assisted by large models. It can flexibly configure dialogue words, dialogue logic, and intelligent engines and supporting models related to dialogue interaction, and configure logical relationships such as branches, loops, and jumps in the flowchart, while completing the configuration of speech recognition, semantic understanding, and speech synthesis engines and models, thereby greatly improving the development efficiency of human-computer dialogue processes. The drag-and-drop design is beneficial for users to efficiently complete the customization of the process, and when encountering complex business scenarios, the definition of the voice robot can also be completed through code configuration, thereby giving the system both ease of use and robustness.

[0164] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for designing a human-computer dialogue process, characterized in that: include: The knowledge base manager configures the knowledge base; The knowledge base is located in the business platform; After the knowledge base is configured, a flowchart is built through the flowchart manager. The flowchart consists of one or more nodes and the logical configuration of the nodes. In the flowchart manager, the nodes of the flowchart are arranged and the logic between the nodes is configured by dragging and dropping. After the flowchart is built, the verification tester performs a verification test on the built flowchart.

2. The method for designing a human-computer dialogue process according to claim 1, characterized in that: in, The knowledge base stores the conversation knowledge background information of the voice robot, which is the knowledge basis for the voice robot to recognize and understand customer conversations. In the subsequent execution of the human-computer dialogue process, the voice robot determines the node jump direction based on the conversation knowledge background information it hits in the knowledge base.

3. The method for designing a human-computer dialogue process according to claim 1, characterized in that: in, The verification tester verifies the syntax and configuration correctness of the flowchart, and conducts a comprehension test after confirming that the syntax and configuration of the flowchart are correct; In the comprehension test, the verification tester simulates a customer call to test the voice robot's intent recognition performance in the human-computer dialogue process. When problems are found in customer intent recognition, the verification tester feeds the problems back to the flowchart manager. The flowchart manager readjusts and builds the flowchart based on the feedback content, thereby optimizing the flowchart.

4. The method for designing a human-computer dialogue process according to claim 1, characterized in that: In the human-computer dialogue process, after the voice robot enters the node of the flowchart, the recognition of the customer's intention includes one or more matching operations performed in sequence, including node branch intention judgment operation, common question answering intention judgment operation, exception handling operation, and fallback operation; Among them, when executing the node branch intention judgment operation, each node is configured with one or more branch routes, each branch route corresponds to a customer intention, when the voice robot recognizes the intention in the branch during the conversation, it means that the intention is hit, the voice robot executes the operation in the node, and jumps out of the current node according to the branch corresponding to the intention, otherwise it enters the common question answering intention judgment operation; Among them, when executing the FAQ intention judgment operation, each node is configured with one or more FAQs. When the corresponding intent is not matched in the node branch, the intent configuration in each FAQ is checked in turn. When a match is found, the FAQ content is broadcasted, otherwise the exception handling operation is entered; Among them, performing abnormal operations includes: Return to the specified node and re-execute the process; or hang up, that is, the voice robot announces the hang-up words and then hangs up the call; or transfer to manual, that is, the voice robot transfers the current call to a manual seat for processing; Among them, anomalies include no matching intention, the customer not speaking, abnormal customer emotions, and abnormal noise; Among them, when the node branch intention judgment operation, the common question answering intention judgment operation, and the exception handling fail to solve the customer needs, the fallback operation is triggered.

5. The method for designing a human-computer dialogue process according to claim 1, characterized in that: The dialogue debugger configures the big model according to the business background of the current human-computer dialogue process; the dialogue debugger inputs the business description and configuration requirements in the business platform into the big model for configuration; The dialogue debugger calls the configured big model to perform big model auxiliary generation operations in the current business context to assist in generating the required content; The dialogue debugger transmits the assisted generated content back to the business platform and stores it in the knowledge base.

6. A human-computer dialogue process design system, characterized in that: Including knowledge base, knowledge base manager, flowchart manager, verification tester; Wherein, the knowledge base manager is used to configure the knowledge base; the knowledge base is located in the business platform; The flowchart manager is used to build a flowchart after the knowledge base configuration is completed. The flowchart consists of one or more nodes and the logical configuration of the nodes. In the flowchart manager, the nodes of the flowchart are arranged and the logic between the nodes is configured by dragging and dropping. Among them, the verification tester is used to perform verification tests on the constructed flowchart after the flowchart is completed.

7. The human-computer dialogue process design system according to claim 6, characterized in that: The knowledge base includes the intention base, interruption base, synonym base, hot word base, FAQ base, and exception handling base. The intent library contains one or more customer intents that the current voice robot needs to understand. The intent is the key meaning of a sentence spoken by the customer. The voice robot associates the customer's speech content with the fixed intent through keyword recognition and regular extraction, thereby identifying and understanding the customer's speech content; The interruption word library includes one or more keywords of the customer's intention to stop the voice robot from continuing to speak. After the voice robot recognizes such keywords, it stops talking to the customer, thereby obtaining a result that the customer interrupts the voice robot in the human-machine dialogue. The synonym database contains one or more corresponding combinations of standard words and synonyms, where one standard word corresponds to one or more synonyms. The voice robot uses the synonyms in the synonym database to deal with the situation where the same intention has multiple expressions in human-computer dialogue. The hot word library contains one or more proper nouns, which are specific proper nouns in a specific business. The voice robot uses the hot words in the hot word library to deal with complex and specific proper nouns in human-computer dialogue. The FAQ database contains one or more FAQs, which are questions frequently asked by customers in the current business and the corresponding voice robot answers. When the voice robot recognizes the customer's intention and matches the answer in the FAQ database, the voice robot performs corresponding dialogue operations according to the configuration. Among them, the exception handling library contains one or more intentions that the voice robot has not recognized. The exception refers to the situation in which the voice robot fails to recognize the intention stored in the knowledge base during the human-computer dialogue, resulting in the inability to continue the process; the exception handling library also contains one or more operations for handling exceptions to ensure that the voice robot fully executes the human-computer dialogue process.

8. The human-computer dialogue process design system according to claim 6, characterized in that: in, The nodes of the flowchart include start node, voice interaction node, transfer to manual node, SMS node, sub-process node, branch node, transfer to external line node, and script node; Among them, the start node is the start node of the human-computer dialogue process. When the start node is executed, the human-computer dialogue process starts; When executing to the voice interaction node, the voice robot broadcasts different words according to the customer's intention; When the call is transferred to a human agent, the voice robot transfers the call to the configured agent group, and the corresponding human agent in the agent group takes over the call. Among them, when executing to the SMS node, the voice robot sends a text message to the customer; When the data interaction node is executed, the voice robot triggers the interface call and sends a data query request to the outside, and replies the queried data information to the customer; Among them, when the needs raised by the customer cannot be solved by a short answer but need to be solved by a short process, the sub-process node is executed to solve the customer's needs through the sub-process; Among them, the branch node is a judgment node. When executing to the branch node, the voice robot determines which branch to execute according to the intention; Among them, when executing to the external line node, the voice robot transfers the call to the external phone; Among them, when executing to the script node, the voice robot executes the configured script code; Among them, when the end node is executed, the human-computer dialogue process ends.

9. The human-computer dialogue process design system according to claim 6, characterized in that: The human-computer dialogue process design system also includes a dialogue debugger, which configures the big model according to the business background of the current human-computer dialogue process; the dialogue debugger inputs the business description and configuration requirements in the business platform into the big model for configuration; The dialogue debugger calls the configured big model to perform big model auxiliary generation operations in the current business context to assist in generating the required content; The dialogue debugger transmits the assisted generated content back to the business platform and stores it in the knowledge base.

10. The human-computer dialogue process design system according to claim 9, It is characterized in that The large model auxiliary operations performed by the dialogue debugger specifically include: The dialogue debugger assists in generating the speech in the voice interaction node. That is, the dialogue debugger generates one or more texts of the speech broadcast by the voice robot based on the current business background and node name using the big model, and stores them in the knowledge base. The dialogue debugger assists in generating keywords in the intent library. That is, the dialogue debugger generates one or more keywords that express specific intents of customers based on the current business context and intent name using the big model, and stores them in the intent library of the knowledge base. The dialogue debugger assists in generating similar questions in the intent library. That is, based on the current business context and intent name, the dialogue debugger uses the big model to generate one or more interrogative sentences that express specific intents of the customer, and stores them in the knowledge base. The dialogue debugger assists in generating common questions and their answers. That is, the dialogue debugger uses a large model to generate one or more questions frequently asked by customers in the current business and their corresponding answers based on the current business context, and stores them in the common questions and answers library of the knowledge base; one common question corresponds to one or more answers.

Images