Dialogue method for node selection based on LLM model, dialogue robot, medium, terminal and program product

Abstract

The invention provides a dialogue method for node selection based on an LLM model, a dialogue robot, a medium, a terminal and a program product, the method is applied to the dialogue robot comprising a plurality of interaction nodes, and the method comprises the following steps: obtaining dialogue content input by a client at an LLM interaction node; outputting the dialogue content and prompt information preset by the LLM interaction node to an LLM model interface, wherein the prompt information comprises a triggering interaction node and a corresponding interaction node triggering path; receiving a node selection result returned by the LLM model interface; and skipping to a corresponding next interaction node based on the node selection result, and executing interaction through the next interaction node. According to the method, the next interaction node of the dialogue process can be accurately selected, and the node selection cost is effectively reduced.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and particularly to a dialogue method, a dialogue robot, a medium, a terminal, and a program product for node selection based on an LLM model. Background Art

[0002] In customer service call scenarios such as logistics, marketing, and finance, traditional robots rely on intent recognition of single-round conversations for subsequent dialogue process management. This mode has the following problems:

[0003] When the currently used single-intent text classification algorithm processes sentences with multiple intents, it may cause semantic understanding ambiguities. To solve this problem, it is necessary to update and iterate the accuracy of the Natural Language Understanding (NLU) model, which depends on the training corpus data of the NLU model. However, in new countries and new scenarios, insufficient data volume of the model may lead to poor recognition effects. At the same time, considering the uncertainty of the product application market, more corpus data cannot be invested, thus causing a contradiction between business development and cost investment.

[0004] When the NLU model cannot solve the problem in time, currently, Frequently Asked Questions (FAQ) can be used as a temporary repair method. However, due to the limitations of the text similarity algorithm, the FAQ may cause other problems, cannot meet all scenario requirements, and it is easy to ignore the impact of the FAQ in subsequent maintenance.

[0005] By analyzing the current user's reply through Natural Language Processing (NLP) technology, the NLU or FAQ is called to obtain the user's intent, thereby generating an intent label. In this way, currently, only a single intent can be generated for the user's reply, lacking the understanding of the context. When calling the intent recognition function according to the intent label at the Condition node of the dialogue process, the true intent of the user cannot be accurately judged, resulting in errors in the Action node selected and executed by the dialogue robot, which is not ideal in some scenarios.

[0006] Therefore, it is necessary to provide a dialogue method, a dialogue robot, a medium, a terminal, and a program product for node selection based on an LLM model to solve the above problems existing in the prior art. Summary of the Invention

[0007] In view of the above-mentioned disadvantages of the prior art, the purpose of the present application is to provide a dialogue method, a dialogue robot, a medium, a terminal, and a program product for node selection based on an LLM model, which are used to solve the technical problem that the prior art cannot accurately select the next node of the dialogue process.

[0008] To achieve the above object and other related objects, a first aspect of the present application provides a dialogue method for node selection based on an LLM model, which is applied to a dialogue robot including multiple interaction nodes, and includes:

[0009] Obtain the dialogue content input by the customer at the LLM interaction node;

[0010] Output the dialogue content and the prompt information preset at the LLM interaction node to the LLM model interface, where the prompt information includes triggerable interaction nodes and their corresponding interaction node trigger paths;

[0011] Receive the node selection result returned by the LLM model interface;

[0012] Based on the node selection result, jump to the corresponding next interaction node and perform the interaction through the next interaction node.

[0013] In some embodiments of the first aspect of the present application, the prompt information further includes recognizable customer intentions; the interaction node trigger path includes the mapping relationship between the customer intentions and the triggerable interaction nodes.

[0014] In some embodiments of the first aspect of the present application, the prompt information further includes scenario prompt information, and the scenario prompt information is used to set the role and task objective of the dialogue robot.

[0015] In some embodiments of the first aspect of the present application, the prompt information further includes triggerable fallback nodes and their corresponding fallback node trigger paths; the fallback node trigger path includes the mapping relationship between the node selection result and the triggerable fallback nodes.

[0016] In some embodiments of the first aspect of the present application, the prompt information further includes triggerable knowledge points; after inputting the dialogue content and the preset prompt information into the LLM model interface and before jumping to the corresponding next interaction node based on the node selection result, it further includes:

[0017] Receive the node selection result and the trigger knowledge points returned by the LLM model interface;

[0018] Perform knowledge point interaction with the customer based on the trigger knowledge points.

[0019] In some embodiments of the first aspect of the present application, after obtaining the conversation content of the customer input at the LLM interaction node, the following steps are further included:

[0020] When the conversation content is blank information, determine whether the LLM interaction node has a preset blank information trigger node;

[0021] If so, jump to the blank information trigger node and perform the interaction through the blank information trigger node;

[0022] If not, output the blank information as the conversation content to the LLM model interface, and perform the jump and interaction of the next interaction node based on the node selection result.

[0023] To achieve the above object and other related objects, the second aspect of the present application provides a dialogue robot for node selection based on the LLM model, including: a plurality of interaction node modules, the interaction node modules include an LLM interaction node module and an NLU interaction node module, the LLM interaction node module is configured to implement the dialogue method as described above, and the NLU interaction node module performs the jump of the interaction node based on the intention recognition result of the customer conversation content.

[0024] To achieve the above object and other related objects, the third aspect of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the method is implemented.

[0025] To achieve the above object and other related objects, the fourth aspect of the present application provides a computer program product, which includes computer program code, and when the computer program code runs on a computer, the computer is enabled to implement the method.

[0026] To achieve the above object and other related objects, the fifth aspect of the present application provides an electronic terminal, including a memory, a processor, and a computer program stored on the memory; the processor executes the computer program to implement the method.

[0027] As described above, the dialogue method, dialogue robot, medium, terminal, and program product for node selection based on the LLM model of the present application have the following beneficial effects:

[0028] Obtain the conversation content of the customer input at the LLM interaction node. By calling the pre-deployed LLM model, based on the current conversation content of the customer input and the context information of the conversation, request the LLM model to make a selection. Input the current conversation content of the customer input, the context information of the conversation, and the prompt information preset by the LLM interaction node into the LLM model interface for the LLM model to make node selection accordingly. Among them, the prompt information includes the triggerable interaction nodes and their corresponding interaction node trigger paths. Then, receive the node selection result returned by the LLM model interface, jump to the corresponding next interaction node based on the node selection result, and directly execute the interaction through the next interaction node, without calling the intent recognition function to generate conditional node judgments, effectively reducing the node selection cost, and each conversation can reduce the communication time and communication cost; and the LLM model makes node selection based on the context information, avoiding the understanding ambiguity that may be caused by the single-intent text classification algorithm when processing sentences with multiple intents, reducing the situation that affects the call quality based on this, and being able to accurately understand the user's intent, making the selected node more appropriate and accurate. Brief Description of the Drawings

[0029] Figure 1 It shows a schematic flow chart of the working principle of a traditional dialogue robot shown as the prior art.

[0030] Figure 2 It shows a schematic flow chart of the dialogue method for node selection based on the LLM model in an embodiment of the present application.

[0031] Figure 3 It shows a schematic framework diagram of the dialogue method for node selection based on the LLM model in an embodiment of the present application.

[0032] Figure 4 It shows a schematic flow chart of the working principle of the dialogue method for node selection based on the LLM model in an embodiment of the present application applied to a voice dialogue robot.

[0033] Figure 5 It shows a schematic flow chart of the working principle of the dialogue method for node selection based on the LLM model in another embodiment of the present application applied to a voice dialogue robot.

[0034] Figure 6 It shows a schematic flow chart of the working principle of the dialogue method for node selection based on the LLM model in an embodiment of the present application applied to a text chat dialogue robot.

[0035] Figure 7 It shows a schematic structural diagram of the dialogue robot for node selection based on the LLM model in an embodiment of the present application.

[0036] Figure 8It shows a schematic structural diagram of an electronic terminal in an embodiment of the present application. Detailed implementation manners

[0037] The following uses specific specific examples to illustrate the implementation manners of the present application. Those skilled in the art can easily understand other advantages and effects of the present application from the content disclosed in this specification. The present application can also be implemented or applied through other different specific implementation manners. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other.

[0038] In the embodiments of the present application, terms such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. For example, the first XX and the second XX are only used to distinguish different XX, and do not limit their sequence. Those skilled in the art can understand that terms such as "first" and "second" do not limit the quantity and execution order, and terms such as "first" and "second" do not necessarily mean different.

[0039] It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" indicate examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the present application should not be interpreted as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, using words such as "exemplary" or "for example" aims to present relevant concepts in a specific manner.

[0040] In the embodiments of the present application, "at least one" means one or more, and "a plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after. "At least one (item)" or its similar expression below refers to any combination of these items, including any combination of single item (item) or plural items (items). For example, at least one (item) of a, b or c can represent: a, b, c, a - b, a - c, b - c or a - b - c, where a, b, c can be single or multiple.

[0041] Before further elaborating on the present invention, the nouns and terms involved in the embodiments of the present invention are described. The nouns and terms involved in the embodiments of the present invention are applicable to the following explanations:

[0042] <1>Traditional dialogue robot: An intelligent dialogue system built based on a series of core technology modules, mainly consisting of a robot composed of an Automatic Speech Recognition (ASR) module, a semantic understanding module of Natural Language Understanding (NLU) / Frequently Asked Questions (FAQ) / Named Entity Recognition (NER), a Dialog Management module, and a response generation module.

[0043] <2>Frequently Asked Questions (FAQ): Refers to specifying certain keywords in the intent as a supplement to the main dialogue process to answer users' frequently asked questions.

[0044] <3>LLM (Large Language Model): A natural language processing model built based on deep learning technology. It is trained with a large amount of text data, can understand and generate human language, and is widely used in tasks such as dialogue systems, text generation, translation, and summarization.

[0045] <4>VoiceBot: Suitable for the application scenario of intelligent customer service with voice calls as the carrier.

[0046] <5>ChatBot: Suitable for the application scenario of intelligent customer service with online chat software as the carrier.

[0047] Figure 1Shows a schematic diagram of the working principle process of a traditional dialogue robot in the prior art. The specific process is as follows: The traditional dialogue robot finishes broadcasting the conversation words and waits for the user to input voice content; the ASR module converts the user's voice input into text; the text converted by the ASR module is input into the NLP model for processing requests, including FAQ2.0 / FAQ1.0 / NLU models; it is judged whether the text converted by the ASR module hits the FAQ2.0 / FAQ1.0 / NLU models. If it hits the corresponding model, the user intention is obtained based on the corresponding model, and the corresponding intention label is generated for output; at the same time, the text converted by the ASR module is also used to query the knowledge base to judge whether it hits the knowledge points; if it hits the knowledge points, it is necessary to judge whether the knowledge base takes precedence over the intention; if the knowledge base takes precedence, the question and answer is based on the knowledge base; if the intention takes precedence, it is necessary to judge whether the intention hits the high-priority trigger point; if the intention hits the high-priority trigger point, the process triggered by the corresponding high-priority trigger point is executed; if the intention does not hit the high-priority trigger point, the intention recognition function is called to judge whether the intention hits the corresponding condition (Condition); if the corresponding condition (Condition) is hit, the condition (Condition) is executed, and if the corresponding condition (Condition) is not hit, it is judged whether the intention hits the low-priority trigger point; if the intention hits the low-priority trigger point, the process triggered by the corresponding low-priority trigger point is executed; if the intention does not hit the low-priority trigger point, the NLU model is processed according to the conversation flow (Story). If the confidence level of the NLU model processing is higher than the threshold, it will jump (Jump) to the corresponding Story according to the current conversation state and the user input; if the confidence level of the NLU model processing is lower than the threshold, it enters the process of whether it hits the low-priority trigger point.

[0048] It can be seen from this that the traditional dialogue robot calls the NLU or FAQ model according to the current user input content to generate intention labels. Based on the generated intention labels, the dialogue robot uses the intention recognition function to judge different conditions, and the dialogue robot executes the output operation according to the judged conditions. This kind of dialogue method lacks the understanding of the context, resulting in errors in the user intention recognized according to the intention label and being unable to accurately select the next node of the dialogue process. Therefore, the present application provides a dialogue method, a dialogue robot, a medium, a terminal and a program product for node selection based on the LLM model, introducing the LLM model for node selection in the dialogue process, writing a preset prompt message (Prompt) description for the possible behavioral intentions of the specified node, and enabling the LLM model to select the subsequent paths within a certain range in combination with the context through the preset Prompt, so as to replace the traditional intention label-oriented robot mode.

[0049] It should be understood that "Story" refers to the dialogue process, specifically the preset capabilities in a scenario. The scenario is modularly split, the user's major intentions are subdivided, and the user is guided into independent scenario modules. Different scenarios have different customized Stories, and it is possible to jump between different stories. For example, in the debt collection scenario, it includes an opening statement, willing to repay, unwilling to repay, already repaid, and seeking help, etc.; in the logistics scenario, it includes asking about the progress, changing the recipient information, returning the logistics as not signed, and querying the logistics location, etc. It should also be understood that the trigger point (Checkpoint, CP) refers to a subprocess that can be triggered at any node in the entire call. The subprocess serves as a supplement to the main process, answers the user's frequent questions, and assists in the normal progress of the main process dialogue. It also supports setting the maximum number of times a global CP can be triggered, and aggregating multiple CPs into a combined CP. Triggering any CP enters the combined CP process, and it can also be divided into high-priority trigger points and low-priority trigger points according to the triggering priority to handle matters with different priorities.

[0050] To facilitate the understanding of the embodiments of the present application, first, in combination with Figure 2 it will be described in detail. Figure 2 FIG. shows a schematic flowchart of a dialogue method for node selection based on an LLM model in an embodiment of the present invention. The dialogue method for node selection based on an LLM model in this embodiment is applied to a dialogue robot including multiple interaction nodes. The method mainly includes the following steps:

[0051] Step S21: Obtain the dialogue content input by the customer at the LLM interaction node.

[0052] The LLM interaction node in this step S21 refers to an interaction node pre-deployed with an LLM model for node selection, and the dialogue content input by the customer is obtained at this LLM interaction node. It should be understood that the interaction node refers to each key link or state in the interaction between the dialogue robot and the customer. Exemplarily, the types of interaction nodes also include start nodes, end nodes, decision nodes, parallel nodes, merge nodes, etc. Among them, the start node is used to start the dialogue process; the end node is used to identify the end of the process; the decision node is used to evaluate conditions and determine the next execution path; the parallel node is used to start multiple parallel tasks and continue the process after all tasks are completed; the merge node is used to wait for all parallel tasks to be completed and then continue a single process.

[0053] In this step S21, the dialogue content input by the customer includes both the dialogue content input by voice and the dialogue content input by text. These two input methods together constitute a diversified interaction channel between the customer and the dialogue robot, thus being applicable to various application scenarios.

[0054] In some embodiments of the present application, the conversation flow in which multiple interaction nodes are located includes a fallback process and a main process with sub-processes inserted therein; the fallback process is used to represent a process that does not belong to the main process.

[0055] It should be understood that the conversation flow formed by connecting multiple interaction nodes in a logical order according to predetermined steps and rules is set for each scenario. The conversation flow includes a main process and a fallback process. Among them, the main process is the main line of the conversation and usually includes multiple steps. The main process supports the insertion of sub-processes to handle specific tasks or complex logics. After the sub-process is completed, it will return to the main process to continue executing the main process, which can be triggered through the trigger point (CP) mechanism. The fallback process is a supplementary mechanism for the main process and is used to handle situations where the main process cannot be executed normally, such as when the user input cannot be understood, variables are missing, etc.

[0056] For ease of understanding, specific examples of the main process, sub-process, and fallback process are given for different fields. For example, in the financial field, the main process is the user account query process, specifically: obtaining the content of the query account input by the user - querying account information - returning the account information query result - ending the conversation; the sub-process is the user identity verification process. In the main process of account query, identity verification is required, specifically: prompting the user to input identity verification information - verifying whether the input identity information is correct - if the verification is passed, continue to execute the main process; the fallback process is the process of not understanding the user input, specifically: when the information input by the user cannot be parsed, prompting the user to input a more specific instruction - if the content that cannot be understood is input multiple times, provide an option for manual transfer. Another example is in the education field. The main process is the user course consultation, specifically: obtaining information such as the content of the consultation course and the class time input by the user - querying the corresponding course information - returning the query result and displaying it to the user - ending the conversation; the sub-process is the course recommendation process. In the main process of consulting courses, asking for recommended courses, specifically: screening out eligible courses - displaying the recommended courses to the user - returning to the main process to continue execution; the fallback process is the technical failure process, specifically: encountering a technical failure in the network connection - notifying the user and providing the estimated recovery time.

[0057] Step S22: Output the conversation content and the prompt information preset by the LLM interaction node to the LLM model interface, where the prompt information includes triggerable interaction nodes and their corresponding interaction node trigger paths.

[0058] Specifically, add a preset prompt message at the LLM interaction node. For example, in the scenario of a logistics customer service call, when there is a poor conversation quality at the node of the signing status, add a preset prompt message for this specific node; call the LLM model pre-deployed at this node to select the subsequent path based on the conversation content and the added preset prompt message. It should be understood that a prompt is an input text provided to the model to guide the model to generate an output conversation text that meets specific scenarios or requirements for the current scenario node.

[0059] In this embodiment, the prompt message includes triggerable interaction nodes and their corresponding interaction node trigger paths. The prompt message of triggerable interaction nodes and their corresponding interaction node trigger paths (Checkpoint Prompt) is used to describe the list of trigger points allowed to be triggered. For example, if the triggerable interaction node is that the user asks for address information, then at this interaction node, the allowed trigger path is Checkpoint_AskAddress (trigger point_asking for address); if the triggerable interaction node is that the user asks for the identity or company name of the caller, or doubts whether it is a fraud call, then at this node, the allowed trigger path is Checkpoint_AskCompany (trigger point_asking for company); if the triggerable interaction node is that the user does not speak, then at this node, the allowed trigger path is Checkpoint_NoVoice (trigger point_scenario of no voice). Output the pre-written triggerable interaction nodes, their corresponding interaction node trigger paths, and the conversation content of the customer to the LLM model interface, and select the next interaction node based on the LLM model, that is, select the subprocess allowed to be triggered by CP.

[0060] In some embodiments of the present application, the prompt message further includes recognizable customer intentions; the interaction node trigger path includes the mapping relationship between the customer intention and the triggerable interaction node. In this embodiment, based on the result of the recognizable customer intention fed back by the LLM model, and node selection is performed according to the recognized customer intention.

[0061] The prompt information (Step Prompt) of recognizable customer intentions is used to describe the customer intentions that need to be recognized. For example, the recognizable customer intention is that the customer has a clear positive response, and based on the LLM model, the next interaction node is selected as Step_conclusionyes (Step_ConclusionYes); the recognizable customer intention is that the customer has a clear negative response, and based on the LLM model, the next interaction node is selected as step_facilitate (Step_Facilitate); the recognizable customer intention is that the user says he didn't hear clearly, and based on the LLM model, the next interaction node is selected as step_ReplyNotClear (Step_ReplyNotClear). The pre-written prompt information of recognizable customer intentions and the customer's conversation content are output to the LLM model interface. Based on the LLM model, the next interaction node is selected, and the next interaction node is triggered according to the recognized customer intention, that is, the next step in the main process is selected.

[0062] In some embodiments of the present application, the prompt information further includes scenario prompt information, and the scenario prompt information is used to set the role and task objectives of the dialogue robot. In this embodiment, for different scenarios, the written scenario prompt information will also be different. Correspondingly, the nodes selected based on the LLM model according to the customer's conversation content are also different.

[0063] The scenario prompt information (System Prompt) is used to describe the background information of the dialogue robot and set the role task objectives of the dialogue robot. The scenario prompt information is maintained by referring to the descriptions in the intention library to set the background and scenario of the conversation, so as to guide the LLM model to generate appropriate responses or perform corresponding actions according to the information such as the role and task objectives of the described dialogue robot. For example, the scenario prompt information is: You (the dialogue robot) are a call service agent of a certain company, and the company is a globally leading smart home electronics brand; the customer calling you needs to be encouraged to pay for the order with a 10% discount coupon on the company's website as soon as possible; try to use effective words to attract the customer and complete their order as much as possible. This example provides a clear background setting, role positioning, task objectives and behavior guidance for the dialogue robot. The pre-written scenario prompt information and the customer's conversation content are input into the LLM model, and the LLM model can generate node selections that conform to the specific situation when processing the customer's conversation content.

[0064] In some embodiments of the present application, the prompt information further includes a triggerable fallback node and its corresponding fallback node trigger path; the fallback node trigger path includes the mapping relationship between the node selection result and the triggerable fallback node. The mapping relationship between the node selection result and the triggerable fallback node is as follows: fallback is performed based on the node selection result fed back by the LLM model; or fallback is performed based on the node selection result fed back by the LLM model and NER; or fallback is performed based on the node selection result fed back by the LLM model and the recognized customer intention; or fallback is performed based on the node selection result fed back by the LLM model, the recognized customer intention, and the current scenario.

[0065] The prompt information of the triggerable fallback node and its corresponding fallback node trigger path (Contigency Prompt) is used to describe the list of fallback processes allowed to be triggered. That is, when the LLM model cannot make a selection within the scope of the Checkpoint Prompt and the StepPrompt, the fallback process is allowed to be triggered. That is to say, based on the LLM model, node selection is performed according to the customer's conversation content and the input triggerable fallback node and its corresponding fallback node trigger path, and the fallback process is triggered according to the node selection result.

[0066] It should be noted that the logical configuration of the fallback node trigger path is to provide a fallback mechanism when the dialogue system encounters situations that cannot be processed, ensuring that the dialogue can continue and reasonable response content can be output. The configuration of the fallback node trigger path includes:

[0067] (1) Select the current node: When the robot fails to understand the user's input, re-enter the current node and ask again in a different way. For example, when the user inputs: "I don't quite understand", the robot replies: "Sorry, I'll ask in a different way: Do you want to know the borrowing amount or the repayment date?".

[0068] (2) Select other steps (Step) of the current dialogue flow (Story): When the user's input is not within the processing scope of the current node but needs to be executed directly, jump to other steps of the current dialogue flow. For example, when the user inputs: "I want to know the status of my logistics", the robot replies: "Okay, your logistics information is as follows:...".

[0069] (3) Select the trigger point (CP): When the user's input is completely unrecognizable (NoRecognize), enter the CP to express that you didn't understand, and then re-enter the step through Action Back to ask again. Exemplarily, when the user inputs: "Whatever", the robot replies: "Sorry, I didn't understand what you meant. Please try again."

[0070] In this step S22, the LLM model is pre-deployed at the LLM interaction node, reducing the demand for corpus. The LLM model is used for node selection in a dedicated context, while avoiding the distortion of the general model in the dedicated environment from affecting the global communication quality.

[0071] Step S23: Receive the node selection result returned by the LLM model interface.

[0072] In this step S23, the result of node selection based on the LLM model is received.

[0073] Step S24: Jump to the corresponding next interaction node based on the node selection result, and perform an interaction through the next interaction node.

[0074] In this step S24, directly jump to the corresponding next interaction node according to the selected node result, and output a reply content through this interaction node to interact with the customer.

[0075] In some embodiments of the present application, the prompt information further includes triggerable knowledge points; after outputting the conversation content and the preset prompt information to the LLM model interface, and before jumping to the corresponding next interaction node based on the node selection result, it further includes: receiving the node selection result and the trigger knowledge point returned by the LLM model interface; performing knowledge point interaction with the customer based on the trigger knowledge point.

[0076] Traditional chatbots first answer from the knowledge base, and then return to the original node to trigger the conversation for the second entry. This method lengthens the conversation time and reduces the user experience. However, in this embodiment, the chatbot that performs node selection based on the LLM model first answers knowledge points, and then directly performs node jumping according to the situation selected by the LLM model, without necessarily jumping back to the original node, saving interaction rounds and enhancing the user experience.

[0077] Exemplarily, the set scenario is that the customer asks how to issue an income certificate, and the customer inputs: "Help me apply for an income certificate". Search in the knowledge base to trigger the knowledge point: "How to issue an income certificate". Interact according to the node selection result and the trigger knowledge point, and output the answer: "The following materials are required to issue an income certificate: ID card, work certificate, salary statement, etc. The specific process can be consulted with the company's human resources department". After answering the knowledge point, directly perform node jumping, which can save interaction rounds.

[0078] In some embodiments of the present application, after obtaining the conversation content of the customer input at the LLM interaction node, the following steps are further included: when the conversation content is blank information, determine whether the LLM interaction node has a preset blank information trigger node; if so, jump to the blank information trigger node and execute the interaction through the blank information trigger node; if not, output the blank information as the conversation content to the LLM model interface, and based on the node selection result, perform the jump and interaction of the next interaction node.

[0079] In this embodiment, in order to avoid the intention generated in the blank information state from affecting the effect of the LLM model through node selection by the LLM model, the above corresponding implementation logic path is designed for the blank information state, which will neither affect the effect of the LLM model nor increase the call time and reduce the node selection cost of the conversation.

[0080] At the interaction node where the LLM model is pre-deployed, there are also configured an interaction node that can be triggered and its corresponding interaction node trigger path, recognizable customer intention, scenario prompt information, a fallback node that can be triggered and its corresponding fallback node trigger path, and prompt information of triggerable knowledge points. That is, at the current interaction node, the LLM model and preset prompt information are configured. According to the conversation content input by the current customer and the context information of the conversation, the LLM model is requested to select a node. The conversation content input by the user, the interaction node that can be triggered and its corresponding interaction node trigger path, recognizable customer intention, scenario prompt information, the fallback node that can be triggered and its corresponding fallback node trigger path, and prompt information of triggerable knowledge points are input into the LLM model for the LLM model to select the next interaction node accordingly. The node selection result returned by the LLM model interface is received, and directly jump to the selected next interaction node based on the node selection result, and output the corresponding reply content through the next interaction node to execute the interaction. By combining the LLM model with preset prompt information and customer conversation content, directly execute the next interaction node without calling the intention recognition function to generate conditional judgment nodes, which greatly reduces the node selection cost and reduces the communication time and communication cost of the conversation.

[0081] Such as Figure 3As shown, assume the scenario is set to inquire about an order, and the customer enters: "Where is the order I placed yesterday?" An LLM model is pre-deployed at the LLM interaction node, and the triggerable interaction node, its corresponding interaction node trigger path, recognizable customer intent, scenario prompt information, and triggerable fallback node, its corresponding fallback node trigger path are configured. For example, the scenario prompt information is specifically: "You are a customer service robot for an e-commerce website, and your task is to help users query the order status." At the current LLM interaction node, the LLM model is called to select a node based on the customer's input conversation content, context information, and preset prompt information. Based on the analysis of the input content by the LLM model, the user's intent to query the order status is recognized, and the corresponding next interaction node, such as the interaction node step_order_status, is selected. The dialogue robot jumps according to the node selected by the LLM model and performs the corresponding operation, that is, queries the order processing system to obtain the latest status of the order, and the output reply content is, for example: "Your order is in transit and is expected to arrive at your address tomorrow. You can track the order status through your account." The dialogue robot based on the LLM model for node selection in this application can directly select a node according to the customer's current input content and the context of the conversation, directly execute the corresponding operation based on the selected node, with a faster response speed and improved user experience.

[0082] In some specific embodiments of the present application, the LLM model is a GPT model. Inputting the customer's current input conversation content and preset prompt information into the GPT model for the GPT model to select an interaction node accordingly includes: performing semantic analysis on the customer's current input conversation content and extracting named entities and context information; based on the named entities, context information, and preset prompt information, recognizing the customer intent based on the GPT model; and selecting a node based on the recognized customer intent based on the GPT model.

[0083] It should be understood that the GPT (Generative Pre-trained Transformer) model is a generative pre-trained language model based on the Transformer architecture, with powerful text generation and understanding capabilities. Through pre-training and fine-tuning, it can be adapted to a variety of natural language processing tasks and is widely used in fields such as dialogue systems, text generation, and translation. The core components of the GPT model include an input representation module, a multi-head attention mechanism, a feed-forward neural network, layer normalization, and a residual connection. Among them, the input representation module converts text into word embeddings and adds positional encoding to preserve word order information; multiple attention heads are used to capture different levels of semantic relationships in the text; the output of the self-attention mechanism is non-linearly transformed based on the feed-forward neural network; layer normalization and residual connections are used to stabilize the training process and accelerate convergence.

[0084] In this embodiment, the GPT model can understand the context information of the dialogue, improve the accuracy of user intention recognition, thereby more effectively manage the dialogue process and enhance the user experience.

[0085] Figure 4 It shows a schematic diagram of the working principle process of the dialogue method for node selection based on the LLM model in an embodiment of the present application applied to a voice dialogue robot. Combining Figure 4Describe in detail the specific workflow of Voice Dialogue Robot 1. After the speech of Voice Dialogue Robot 1 ends, it waits for the user to input voice content. It judges whether subsequent actions are predefined. If subsequent actions are defined, Voice Dialogue Robot 1 executes the subsequent actions. If subsequent actions are not defined, it judges whether the pre-deployed LLM model is turned on. If it is not turned on, it enters the preset connection node to continue executing subsequent decisions. If it is in the on state, it judges whether the content of the voice input is empty. If it is not in the empty state, it requests the knowledge base for knowledge point Q&A. If the voice input is unclear, interfered by background noise, the speech content exceeds the understanding range of the ASR module, etc., that is, the content of the voice input is empty, or after the knowledge base Q&A ends, it enters the judgment of whether the trigger point of the current node reaches the set maximum trigger times. If the set maximum trigger times are reached, only the Prompt of this node is given, and the LLM model is requested to select. If the set maximum trigger times are not reached, the Prompt of this node and the trigger point prompt words that can be triggered are given, and the LLM model is requested to select. Then, it enters the judgment of whether the model hits the knowledge base. If it hits the knowledge base, Q&A is performed according to the knowledge base. If it does not hit the knowledge base, it judges whether the node selected by the LLM model is appropriate. If the node selected by the LLM model is appropriate, it judges whether the node selected by the LLM model is a trigger point. If it is a trigger point, the sub-process corresponding to this trigger point is executed. If it is not a trigger point, it jumps to the steps of the main process and executes the corresponding steps of the main process. If the LLM model does not select an appropriate node, the fallback logic is triggered; and it judges whether the fallback logic is configured as a trigger point. If it is a trigger point, the trigger point is triggered. If it is not a trigger point, it jumps to the steps and executes the corresponding steps of the main process.

[0086] In this embodiment, after the user ends the call, at this current node, four types of information, namely all the content of the human-machine dialogue, the configured path of the current node, the trigger interaction nodes and descriptions allowed to be triggered by the current node, and the knowledge points allowed to be triggered by the current node, are input into the LLM model for selection. The LLM model will return two results: whether it hits the knowledge point and the selected path. The dialogue robot will perform the following actions according to the returned results: If it does not hit the knowledge point, it jumps to the node of the selected path; if it hits the knowledge point, it will first answer the knowledge point and then jump to the node of the selected path.

[0087] The dialogue robot that selects nodes based on the LLM model can correct some scenarios where the ASR module recognizes errors, such as partial words being omitted or incorrect in the text recognized by the ASR module, but it cannot correct the recognition errors of slot value data, such as incorrect recognition of the express waybill number.

[0088] The return results of the dialogue robot are as follows: If the recognition result of the returned language is obtained, the language variable is assigned a value, and the corresponding engine switch is performed according to the configuration; if the filling result of the NER model slot is returned, the slot filling operation is performed normally; if the hit knowledge point is returned, the knowledge base answer is given. After the answer is completed, it will not directly return to the original node, but directly jump according to the node selected by the LLM model; jump according to the returned node. If the LLM model service is abnormal or times out, the fallback process is to enter the current node again. After the same node has three consecutive exceptions, the fallback statement is broadcast and then the call is hung up, and an alarm notification is also sent. It should be understood that the slot value acts as a global attribute field throughout the call process. The slot value needs to be declared and extracted at the beginning of the dialogue process, and the slot value can be obtained through the getslot method throughout the call cycle. A slot is a variable in the dialogue system used to store specific information, usually related to the user's intention or task. The slot value is the specific value of the slot, usually obtained through user input or system inference. The NER model supports filling slot values such as repayment methods, repayment times, and reasons for non-repayment in the collection scenario; it supports filling slot values such as waybill numbers and sender names in the logistics scenario, and each slot corresponds to a preset slot value.

[0089] After a round of call by the dialogue robot ends, the following call data is fed back for subsequent adjustment of the robot dialogue management process: Record the step or trigger point name given by the LLM model after the current node requests the LLM model to select a node; record whether the node currently selected by the LLM model is a fallback node; record the total number of Tokens (tokenization units) output by the LLM model and the total number of input Tokens for the entire call.

[0090] Figure 5 Shown is a schematic diagram of the working principle process of the dialogue method for node selection based on the LLM model in another embodiment of the present application applied to a voice dialogue robot. Combining Figure 5Describe in detail the specific working process of the voice dialogue robot 2. After the voice dialogue robot 2 finishes speaking the dialogue, it waits for the user to input voice content; determines whether subsequent actions are predefined. If subsequent actions are defined, the voice dialogue robot 2 executes the subsequent actions; if subsequent actions are not defined, it determines whether the pre-deployed LLM model is turned on. If it is not turned on, it enters the preset connection node to continue executing subsequent decisions. If it is in the on state, it determines whether the content of the voice input is empty. When the content of the voice input is empty, it enters the execution path of the NoVoice state, specifically: generates the system intention of NoVoice, and determines whether the current node trigger point has reached the set maximum trigger times. If the maximum trigger times have not been reached, it determines whether the trigger point list that can be triggered contains the system intention of this NoVoice. If the system intention of this NoVoice is in the trigger point list, it executes the sub-process; if the system intention of this NoVoice is not in the trigger point list, or the maximum trigger times have been reached, it determines whether the LLM model contains the system intention. If the LLM model contains the system intention, it jumps to the steps of the main process. If the LLM model does not contain the system intention, it triggers the fallback node and determines whether the fallback node is configured as a trigger point. If it is configured as a trigger point, it executes the sub-process. If it is not configured as a trigger point, it jumps to the steps of the main process and executes the corresponding steps in the main process. When the content of the voice input is not empty, it enters the execution path of requesting knowledge base Q&A. After the knowledge base Q&A is completed, it enters the path of requesting the LLM model for node selection, which is the same as the judgment path of the above-mentioned voice dialogue robot 1 and will not be elaborated here.

[0091] It should be understood that the NoVoice state is used to identify the situation where the user does not speak, or the voice input cannot be recognized and there is only noise, that is, the dialogue content is blank information.

[0092] In this embodiment, in order to avoid the system intention generated by the NoVoice state from affecting the effect of the LLM model through node selection of the LLM model, the above corresponding implementation logic path is designed for the NoVoice state, which will neither affect the effect of the LLM model nor increase the call time and reduce the node selection cost of the dialogue.

[0093] Figure 6 It shows a schematic diagram of the working principle process of the dialogue method for node selection based on the LLM model in an embodiment of the present application applied to a text chat dialogue robot. Combined with Figure 6Describe in detail the specific working process of the text chatbot. The text chatbot sends a conversation prompt and waits for the user to input text. It determines whether subsequent actions are predefined based on the current entry count. If subsequent actions are defined, the text chatbot executes them. If no subsequent actions are defined, it checks whether the pre-deployed LLM model is enabled. If not, it enters a preset connection node to continue subsequent decision-making. If it is enabled, it checks whether a silent wake-up mechanism is configured, whether an associated resource intent is triggered, or whether it is text information. If a silent wake-up mechanism is configured, an associated resource intent is triggered, or it is not text information, generating a system intent of NoVoice, or generating a system intent of an associated resource intent, or generating a system intent of ReplyTypeNoRecognize, all enter the next judgment and execution path, specifically: it determines whether the current node trigger point has reached the set maximum trigger count. If it has not reached the maximum trigger count, it checks whether the list of trigger points that can be triggered contains the above three system intents. If the above three system intents are in the trigger point list, it executes a sub-process. If the above three system intents are not in the trigger point list, it continues to check whether it is a system intent of NoVoice. If it is a system intent of NoVoice, it copies the node configuration and executes the silent wake-up conversation prompt. If it is not a system intent of NoVoice, or if the maximum trigger count has been reached, it checks whether the above three system intents are in the LLM model. If the LLM model has this scenario prompt information, it jumps to the steps of the main process. If the LLM model does not have this scenario prompt information, it triggers a fallback node and checks whether the fallback node is configured as a trigger point. If it is configured as a trigger point, it executes a sub-process. If it is not configured as a trigger point, it jumps to the steps of the main process and executes the corresponding steps in the main process. If the input conversation content is text information, it enters the execution path of requesting knowledge base Q&A. After the knowledge base Q&A is completed, it enters the path of requesting the LLM model for node selection, which is the same as the judgment path of the above voice chatbot 1 and will not be elaborated here.

[0094] It should be understood that the ReplyTypeNoRecognize state refers to the recognized state when input is made through methods such as pictures, files, audio, and video, that is, a state that does not belong to text information.

[0095] In this embodiment, the above corresponding execution logic paths are designed for the silent wake-up state and the ReplyTypeNoRecognize state, enabling the text chatbot to remain active under special circumstances and enhancing the robustness and user experience of the system.

[0096] Figure 7 It is a schematic block diagram of a chatbot for node selection based on the LLM model provided by an embodiment of the present application. As Figure 7As shown in the figure, the dialogue robot 700 for node selection based on the LLM model includes: a plurality of interaction node modules, the interaction node modules include an LLM interaction node module 701 and an NLU interaction node module 702. The LLM interaction node module is configured to implement the dialogue method as described above, and the NLU interaction node module performs the jump of the interaction node based on the intention recognition result of the customer dialogue content.

[0097] It should be noted that the dialogue method for node selection based on the LLM model of the present application can be nested into a robot guided by intention tags based on NLU / FAQ, that is, the method of the present application can be used for node selection for a certain node, while the original scheme is still used for the remaining nodes.

[0098] In this embodiment, the LLM interaction node module 701 reuses the NLU intention library and the FAQ database of the NLU interaction node module.

[0099] It should be understood that the specific processes for each module to execute the corresponding steps above have been described in detail in the above method embodiments. For the sake of brevity, they will not be repeated here.

[0100] It should also be understood that the division of modules in the embodiments of the present application is illustrative, only a logical function division. In actual implementation, there may be other division methods. In addition, in each embodiment of the present application, the functional modules can be integrated in one processor, or can exist separately physically, or two or more modules can be integrated in one module. The above integrated modules can be implemented in the form of hardware or in the form of software functional modules.

[0101] Figure 8 is a schematic block diagram of an electronic terminal provided by an embodiment of the present application. As Figure 8 shown, the electronic terminal 800 includes: at least one processor 801, a memory 802, at least one network interface 803, and a user interface 805. Each component in the electronic terminal 800 is coupled together through a bus system 804. It can be understood that the bus system 804 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 804 also includes a power bus, a control bus, and a status signal bus. However, for the sake of clear illustration, in Figure 8 all kinds of buses are labeled as the bus system.

[0102] Among them, the user interface 805 may include a display, a keyboard, a mouse, a trackball, a click gun, a button, a button, a touchpad, or a touch screen, etc.

[0103] It can be understood that the memory 802 can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include but not limited to these and any other suitable categories of memories.

[0104] The memory 802 in the embodiments of the present invention is used to store various categories of data to support the operation of the electronic terminal 800. Examples of such data include: any executable programs for operating on the electronic terminal 800, such as the operating system 8021 and application programs 8022; the operating system 8021 contains various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application programs 8022 can include various application programs, such as a media player, a browser, etc., for implementing various application services. Implementing the XX method provided by the embodiments of the present invention can be included in the application programs 8022.

[0105] The method disclosed in the above embodiments of the present invention can be applied to or implemented by the processor 801. The processor 801 may be an integrated circuit chip with signal processing capabilities. In the implementation process, the steps of the above method can be completed by the integrated logic circuit in the hardware of the processor 801 or instructions in software form. The above processor 801 can be 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 processor 801 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor 801 can be a microprocessor or any conventional processor, etc. Combining the steps of the accessory optimization method provided by the embodiments of the present invention can be directly embodied as being completed by the hardware decoding processor, or by a combination of the hardware and software modules in the decoding processor. The software module can be located in a storage medium, and this storage medium is located in the memory. The processor reads the information in the memory and combines its hardware to complete the steps of the foregoing method.

[0106] In an exemplary embodiment, the electronic terminal 800 may be implemented by one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) for performing the foregoing method.

[0107] According to the method provided by the embodiments of the present application, the present application further provides a computer program product, which includes: computer program code that, when running on a computer, causes the computer to execute Figures 2 to 6 the method of any one of the illustrated embodiments.

[0108] According to the method provided by the embodiments of the present application, the present application further provides a computer-readable storage medium storing program code that, when running on a computer, causes the computer to execute Figures 2 to 6 the method of any one of the illustrated embodiments.

[0109] The terms "component", "module", "system", etc. used in this specification are used to denote computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component may be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, and / or a computer. By way of illustration, both an application running on a computing device and the computing device can be components. One or more components may reside in a process and / or execution thread, and a component may be located on one computer and / or distributed between two or more computers. In addition, these components may execute from various computer-readable media storing various data structures. A component may communicate, for example, through local and / or remote processes via signals having one or more data packets (e.g., data from two components interacting with another component in a local system, a distributed system, and / or a network, such as data interacting with other systems via signals on the Internet).

[0110] Those of ordinary skill in the art will appreciate that the various illustrative logical blocks and steps described in connection with the embodiments disclosed herein can be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled artisans may use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.

[0111] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0112] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only a logical function division, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical, or other forms.

[0113] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0114] In addition, the functional units in each embodiment of this application can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit.

[0115] In the above embodiments, the functions of each functional unit can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more integrated available media. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a magnetic tape), an optical medium (for example, a high-definition digital video disc (DVD)), or a semiconductor medium (for example, a solid state disk (SSD), etc.).

[0116] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs, etc., which can store program codes of various types.

[0117] As described above, the above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all of them should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0118] In summary, the present application provides a dialogue method, a dialogue robot, a medium, a terminal, and a program product for node selection based on an LLM model. The dialogue content input by the customer is obtained at the LLM interaction node. By calling the pre-deployed LLM model, the LLM model is requested to make a selection according to the current dialogue content input by the customer and the context information of the dialogue. The current dialogue content input by the customer, the context information of the dialogue, and the prompt information preset by the LLM interaction node are input into the LLM model interface for the LLM model to make a node selection based on this. The prompt information includes triggerable interaction nodes and their corresponding interaction node trigger paths. Then, the node selection result returned by the LLM model interface is received, and the corresponding next interaction node is jumped to based on the node selection result, and the interaction is directly executed through the next interaction node, without calling the intent recognition function to generate conditional node judgments, effectively reducing the node selection cost, and reducing the communication time and communication cost for each conversation; and the LLM model makes a node selection based on the context information, avoiding the understanding ambiguity that may be caused by the single-intent text classification algorithm when processing sentences with multiple intents, reducing the occurrence of situations affecting the call quality based on this, and being able to accurately understand the user's intent, making the selected nodes more appropriate and accurate. Therefore, the present application effectively overcomes various disadvantages in the prior art and has high industrial utilization value.

[0119] The above embodiments are only illustrative of the principles and effects of the present application and are not intended to limit the present application. Any person familiar with this technology can modify or change the above embodiments without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by those with ordinary knowledge in the technical field without departing from the spirit and technical ideas disclosed by the present application should still be covered by the claims of the present application.

Claims

1. A dialog method for node selection based on the LLM model, characterized in that: Applied to a conversational robot including a plurality of interaction nodes, the method comprises: Obtain the conversation content entered by the customer at the LLM interaction node; Outputting the conversation content and the preset prompt information of the LLM interaction node to the LLM model interface, wherein the prompt information includes the triggerable interaction node and its corresponding interaction node trigger path; Receive the node selection result returned by the LLM model interface; Jump to the corresponding next interaction node based on the node selection result, and perform interaction through the next interaction node.

2. The dialog method for node selection based on the LLM model according to claim 1, characterized in that: The prompt information also includes a recognizable customer intention; the interaction node triggering path includes a mapping relationship between the customer intention and the triggerable interaction node.

3. The dialog method for node selection based on the LLM model according to claim 1, characterized in that: The prompt information also includes scenario prompt information, and the scenario prompt information is used to set the role and task objectives of the dialogue robot.

4. The dialog method for node selection based on the LLM model according to claim 1, characterized in that: The prompt information also includes a triggerable fallback node and its corresponding fallback node triggering path; the fallback node triggering path includes a mapping relationship between the node selection result and the triggerable fallback node.

5. The dialog method for node selection based on the LLM model according to claim 1, characterized in that: The prompt information also includes triggerable knowledge points; After the dialog content and the preset prompt information are output to the LLM model interface, and before the node selection result is jumped to the corresponding next interaction node, the method further includes: Receive the node selection result and trigger knowledge point returned by the LLM model interface; Knowledge point interaction is performed with the customer based on the triggering knowledge point.

6. The dialog method for node selection based on the LLM model according to any one of claims 1 to 5, characterized in that: After the LLM interaction node obtains the conversation content input by the customer, the method further includes: When the conversation content is blank information, determining whether the LLM interaction node has a preset blank information trigger node; If yes, jump to the blank information trigger node and perform interaction through the blank information trigger node; If not, the blank information is output as the dialogue content to the LLM model interface, and the jump and interaction of the next interactive node are executed based on the node selection result.

7. A conversational robot for node selection based on the LLM model, characterized in that: include: A plurality of interaction node modules, wherein the interaction nodes include an LLM interaction node module and an NLU interaction node module, wherein the LLM interaction node module is configured to implement the conversation method described in any one of claims 1 to 6, and wherein the NLU interaction node module executes the jump of the interaction node based on the intention recognition result of the customer conversation content.

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

9. A computer program product, characterized in that The computer program product includes computer program codes, and when the computer program codes are executed on a computer, the computer is enabled to implement the method according to any one of claims 1 to 6.

10. An electronic terminal comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the method according to any one of claims 1 to 6.

Images