Dialogue method, device, system, equipment, medium, program product and vehicle

Abstract

The invention discloses a dialogue method, device, system and equipment, a medium, a program product and a vehicle, and belongs to the technical field of natural language processing. The method comprises the following steps: receiving a dialogue request; determining a dialogue backtracking strategy according to the dialogue request and stored dialogue key information; the dialogue key information comprises dialogue content, dialogue content integrity and dialogue progress information; and outputting a target reply content corresponding to the dialogue request according to the dialogue backtracking strategy. When the dialogue request is received, the dialogue backtracking strategy is determined according to the stored dialogue key information, dialogue backtracking can be carried out according to the stored dialogue key information, the possibility of obtaining reply content again is provided for a user, and the dialogue experience of the user is improved.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and specifically to a dialogue method, apparatus, system, device, medium, program product and vehicle. Background Art

[0002] At present, the car machine has realized the function of large model question and answer, which is based on a single round of interaction with users in the form of cards, answering users' questions, and bringing more help and better experience to users. Generally speaking, the ideal interaction process is that the user asks a question, and the voice assistant reads the answer while displaying the text on the card. The single round of large model question and answer cards will count down and exit at the end of the voice broadcast, marking the end of this round of dialogue. After the dialogue ends, the dialogue information of this round will be discarded, and it will not be stored and will not support the backtracking function. If the card exits due to other reasons during the voice broadcast, for example, the user actively exits the card because of something temporarily, or the card is exited by other services, or even accidentally touches the card, the user will not be able to read or listen to all the content on the current card. Because the response length of the current large model is relatively long, this phenomenon of exiting midway is very common.

[0003] After the large model question and answer card disappears, if the user wants to continue to learn about the previous topic, he can only ask the same question again. However, due to the problem of the large model generation mechanism, the user cannot get a completely consistent answer and must listen to the answer again from the beginning. For such a scenario, there is currently no good solution to provide users with conversation backtracking. A disappeared card cannot be displayed again, and the half of the answer that was not heard last time will never be traced. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a conversation method, apparatus, system, device, medium, program product and vehicle to solve the deficiency in the prior art that the reply content cannot be obtained again after the large model conversation is exited midway.

[0005] In order to achieve the above-mentioned object, the first aspect of the present application provides a dialogue method, the dialogue method comprising: Receive a conversation request; Determine a conversation backtracking strategy according to the conversation request and the stored conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; According to the dialogue backtracking strategy, the target reply content corresponding to the dialogue request is output.

[0006] According to the above technical means, when a conversation request is received, a conversation backtracking strategy is determined based on the stored conversation key information, and the conversation can be backtracked based on the stored conversation key information, providing the user with the possibility of obtaining the reply content again, thereby improving the user's conversation experience.

[0007] In some feasible embodiments, the conversation content includes first reply content; and determining the conversation backtracking strategy according to the conversation request and the stored conversation key information includes: Determine whether there is key information of the conversation to determine whether there is conversation content for backtracking; Determining whether the first reply content has been generated according to the completeness of the conversation content; Determining whether the first reply content has been outputted according to the conversation progress information; A corresponding conversation backtracking strategy is determined according to whether there is conversation content for backtracking, whether the first reply content has been generated, and whether the first reply content has been output.

[0008] According to the above technical means, whether there is key information of the conversation indicates whether there is conversation content for backtracking, the completeness of the conversation content indicates whether the first reply content has been generated, and the conversation progress information indicates whether the first reply content has been output. Based on this, the status of the previous round of conversation can be analyzed, which provides a basis for determining the conversation backtracking strategy.

[0009] In some feasible embodiments, according to whether there is a conversation content for backtracking, whether the first reply content is completely generated, and whether the first reply content is completely output, determining a corresponding conversation backtracking strategy includes: If there is no conversation content for backtracking, determining and executing a first conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated and the first reply content has been output, then determining and executing a second conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated but the first reply content has not been output, determining and executing a third conversation backtracking strategy; If there is conversation content for backtracking, and the first reply content has not been completely generated, a fourth conversation backtracking strategy is determined and executed.

[0010] According to the above technical means, based on the status of the previous round of conversation obtained by analyzing the key information of the conversation, the actual needs of the user to trigger conversation backtracking are clarified according to the status of the previous round of conversation, and different conversation backtracking strategies are executed according to different needs to maximize the user's conversation backtracking needs and improve the user's conversation experience.

[0011] In some feasible embodiments, the key conversation information is stored in the following manner: When the reply content of the current round of dialogue begins to be generated, the key dialogue information of the previous round of dialogue in the cache is cleared, and the key dialogue information of the current round of dialogue is stored in the cache.

[0012] According to the above technical means, the key information of the conversation is cleared when the reply content begins to be generated. If the conversation is backtracked, the conversation questions and reply content of the previous round of conversation in the key information have been obtained at this time. Clearing the key information of the conversation will not affect the conversation backtracking, and it can also ensure that the key information of the current round of conversation can be accurately stored.

[0013] In some feasible embodiments, the conversation backtracking strategy includes: a first conversation backtracking strategy, a second conversation backtracking strategy, a third conversation backtracking strategy, and a fourth conversation backtracking strategy; The first conversation backtracking strategy includes: starting a new conversation; The second dialogue backtracking strategy includes: obtaining the dialogue question and the first reply content of the previous round from the dialogue key information; Generate supplementary reply content according to the dialogue question and the first reply content; The third conversation backtracking strategy includes: obtaining the first reply content and conversation progress information from the conversation key information; The fourth dialogue backtracking strategy includes: The conversation questions of the previous round are obtained from the conversation key information, and new reply content is generated according to the conversation questions.

[0014] According to the above technical means, when a new conversation is started when there is no conversation content for backtracking, the user's conversation needs can be responded to, and all requests must be responded to; when the first reply content is generated and output, it is determined that the user has consumed the content of the previous round of conversation. At this time, the purpose of the user triggering conversation backtracking is to obtain more content based on the previous question, so the reply content is supplemented in the subsequent conversation to provide the user with more required content and improve the user experience; when the first reply content is complete, but the first reply content has not been output, it is determined that the user needs to continue to receive the same content as before. At this time, the first reply content and the conversation progress information are obtained, and the content to be output after the strategy is executed is prepared, so that the user can continue to read or listen to the answer that was read or listened to half of the last conversation; when the first reply content for backtracking is incomplete, new reply content is generated according to the conversation question, and a complete answer can be provided for the conversation question that the user did not get an answer to before, thereby improving the user experience.

[0015] In some feasible embodiments, the conversation backtracking strategy includes: a first conversation backtracking strategy, a second conversation backtracking strategy, a third conversation backtracking strategy, and a fourth conversation backtracking strategy; After determining the conversation backtracking strategy, when the conversation backtracking strategy is the first conversation backtracking strategy, storing the conversation key information of the new conversation in the cache; When the conversation backtracking strategy is the second conversation backtracking strategy, when starting to generate the reply supplementary content, clear the conversation key information of the previous round of conversation in the cache, and store the conversation request and the generated reply supplementary content in the cache; When the conversation backtracking strategy is the third conversation backtracking strategy, after obtaining the first reply content from the conversation key information, clearing the conversation key information of the previous round of conversation in the cache, storing the conversation request in the cache, and transferring the first reply content as the target reply content to the cache; When the conversation backtracking strategy is the fourth conversation backtracking strategy, when starting to generate new reply content, the conversation key information of the previous round of conversation in the cache is cleared, and the conversation request and the generated new reply content are stored in the cache.

[0016] According to the above technical means, different conversation backtracking strategies correspond to different storage methods. When the conversation backtracking strategy needs to be generated, the key conversation information of the previous round of conversation in the cache is cleared when the generation begins, and the conversation request and the generated content are stored in the cache to ensure that the key conversation information is updated in real time. When the conversation backtracking strategy needs to backtrack and output the reply content of the previous round, the first reply content is transferred to the cache as the target reply content to ensure that the key conversation information can be stored normally, while also ensuring the accuracy of the key conversation information.

[0017] In some feasible embodiments, outputting target reply content corresponding to the dialogue request according to the dialogue backtracking strategy includes: Execute the dialogue backtracking strategy to obtain the target reply content; According to the target reply content, outputting the target reply content based on an output strategy; The output strategy includes a voice broadcast strategy and a reply content display strategy.

[0018] According to the above technical means, the corresponding target reply content is obtained after executing the dialogue backtracking strategy, and then based on the output strategy output, different target reply contents are output according to different user needs to improve the user dialogue experience.

[0019] In some feasible embodiments, the voice broadcast strategy includes a guide information output and a target reply content output, and the guide information output and the target reply content output are linked by monitoring a guide information output end signal; The guidance information includes a new dialogue start prompt voice when the first dialogue backtracking strategy is executed, and an original dialogue continuation prompt voice when the third dialogue backtracking strategy or the fourth dialogue backtracking strategy is executed; The target reply content output includes: outputting the reply content of the new dialogue after executing the first dialogue backtracking strategy; or outputting the supplementary content of the reply after executing the second dialogue backtracking strategy; or after executing the third conversation backtracking strategy, continuing to output the first reply content based on the conversation progress information; Or the output of the new reply content after executing the fourth dialogue backtracking strategy.

[0020] According to the above-mentioned technical means, the guidance information can play a connecting role in the conversation process, prompting the user the direction of the next conversation. After listening to the end signal of the guidance information output, the execution result is output, which can avoid the execution result output interrupting the guidance information output, thereby improving the integrity of the user's listening content. The output of the target reply content adopts different output methods according to whether the reply content has been output. In particular, for the reply content that has been partially output in the previous round, the method of continuing to output according to the conversation progress information is adopted, so that the user can continue to obtain the reply content that was half obtained in the previous round of conversation, thereby improving the efficiency and experience of obtaining reply content in the user conversation.

[0021] In some feasible embodiments, after receiving the dialog request and before determining the dialog backtracking strategy, the method further includes: identifying the intent of the conversation request; Determine whether the intention is the intention of the target vertical domain; When the intention is the intention of the target vertical domain, determining whether the dialog request triggers dialog backtracking; When the dialog backtracking is triggered, the dialog backtracking strategy determination step is executed.

[0022] According to the above technical means, after receiving a conversation request, it can be determined whether to trigger conversation backtracking based on the identified intention, so that when conversation backtracking is triggered, the step of determining the conversation backtracking strategy is executed. This can avoid executing the step of determining the conversation backtracking strategy when the conversation request does not trigger conversation backtracking, thereby occupying system memory.

[0023] In some feasible embodiments, the dialogue method further includes: When a conversation interruption signal is received, the interruption signal is synchronized to the voice image, and the voice image performs a backtracking prompt.

[0024] According to the above technical means, when a conversation is interrupted, prompting through voice images can prompt the user to recall the conversation before the interruption, so that the user can trace back the conversation in time and obtain the content he wants.

[0025] A second aspect of the present application provides a dialogue device, the dialogue device comprising: A data receiving unit, used for receiving a dialogue request; A strategy determination unit, configured to determine a conversation backtracking strategy according to the conversation request and stored conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; The execution unit is used to output the target reply content corresponding to the dialogue request according to the dialogue backtracking strategy.

[0026] According to the above-mentioned technical means, when the data receiving unit of the device receives a conversation request, the strategy determination unit determines the conversation backtracking strategy based on the stored conversation key information, and can perform conversation backtracking based on the stored conversation key information, providing the user with the possibility of obtaining the reply content again, thereby improving the user's conversation experience.

[0027] A third aspect of the present application provides a dialogue system, the system comprising: Natural language understanding module, chat application module, large model module, cache and output module; The natural language understanding module is used to understand the dialogue request and send the understood user intention to the chat application module; The chat application module is used to query the stored conversation key information from the cache according to the user intention, and determine the conversation backtracking strategy according to the conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; The chat application module is further used to call the large model module and / or the natural language understanding module according to the dialogue backtracking strategy to determine the target reply content corresponding to the dialogue request; The output module is used to receive the target reply content sent by the chat application module and output the target reply content.

[0028] According to the above-mentioned technical means, the natural language understanding module in the system receives and recognizes conversation requests, and the chat application module determines the conversation backtracking strategy based on the stored conversation key information. The conversation can be backtracked based on the stored conversation key information, providing users with the possibility of obtaining reply content again, thereby improving the user conversation experience.

[0029] In some feasible embodiments, the output module includes a voice broadcast module and a graphical interface module; If the conversation backtracking strategy determined by the chat application module is the second conversation backtracking strategy, the chat application module requests the natural language understanding module to continue generating semantics, the natural language understanding module identifies the semantics of the previous round of conversation questions and sends it to the large model module, the large model module generates reply supplementary content according to the semantics of the previous round of conversation questions and sends it to the chat application module, the chat application module sends the reply supplementary content to the graphical interface module for display, and at the same time sends the reply supplementary content to the voice broadcast module for broadcasting; If the conversation backtracking strategy determined by the chat application module is the third conversation backtracking strategy, the chat application module obtains the first reply content as the target reply content, and sends it to the graphical interface module for display, and at the same time sends it to the voice broadcast module for voice broadcasting of the target reply content based on the conversation progress information; If the conversation backtracking strategy determined by the chat application module is the fourth conversation backtracking strategy, the chat application module requests the natural language understanding module for the previous round of conversation questions, the natural language understanding module identifies the semantics of the previous round of conversation questions and sends them to the large model module, the large model module generates new reply content according to the semantics of the previous round of conversation questions and sends it to the chat application module, the chat application module sends the new reply content to the graphical interface module for display, and at the same time sends the new reply content to the voice broadcast module for broadcast.

[0030] According to the above technical means, different reply contents are generated according to different conversation backtracking strategies and displayed and outputted. According to the overall situation of the user's previous round of conversation, the reply contents can be outputted in a targeted manner to improve the user's conversation experience.

[0031] A fourth aspect of the present application provides an electronic device, including: a memory configured to store instructions; and A processor is configured to call the instructions from the memory and implement the dialog method when executing the instructions.

[0032] A fifth aspect of the present application provides a vehicle, including a control system, which applies the dialogue method.

[0033] A sixth aspect of the present application provides a machine-readable storage medium, on which instructions are stored, and the instructions are used to enable a machine to execute the described dialogue method.

[0034] A seventh aspect of the present application provides a computer program product, comprising a computer program, which implements the dialogue method when executed by a processor.

[0035] Through the above technical solution, a conversation method is provided. When receiving a conversation request, the method determines a conversation backtracking strategy according to the stored conversation key information. The conversation can be backtracked according to the stored conversation key information, providing the user with the possibility of obtaining the reply content again, thereby improving the user's conversation experience.

[0036] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present application, but do not constitute a limitation on the embodiments of the present application. In the accompanying drawings: Figure 1 A schematic diagram of a process flow of a dialogue method according to an embodiment of the present application is schematically shown; Figure 2 A schematic diagram of a triggering process of a dialogue method according to an embodiment of the present application is shown schematically; Figure 3 A schematic diagram of a dialogue method execution logic according to an embodiment of the present application is shown; Figure 4 A block diagram of a dialogue system according to an embodiment of the present application is schematically shown; Figure 5 The first backtracking strategy interaction diagram of the dialogue method according to the embodiment of the present application is schematically shown; Figure 6 The second backtracking strategy interaction diagram of the dialogue method according to the embodiment of the present application is schematically shown; Figure 7 The third backtracking strategy interaction diagram of the dialogue method according to the embodiment of the present application is schematically shown; Figure 8 The fourth backtracking strategy interaction diagram of the dialogue method according to an embodiment of the present application is schematically shown. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the specific implementation methods described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0039] It should be noted that if the embodiments of the present application involve directional indications (such as up, down, left, right, front, back...), such directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0040] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present application, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0041] Terminology explanation: NLU (Natural Language Understanding): Natural language understanding. Convert the user's natural language input into a representation that the computer can understand.

[0042] DM (Dialog Manager): Dialogue management, a module responsible for handling the dialogue process in the intelligent voice interaction system, parsing the user's input and determining the next step of the dialogue process.

[0043] NLG (Natural Language Generation): Natural language generation refers to the ability of intelligent dialogue systems to automatically generate natural language text that conforms to grammatical and semantic rules to respond to user input.

[0044] Figure 1 The flowchart of a dialogue method according to an embodiment of the present application is schematically shown. The method can be applied to terminal devices, such as car computers, smart phones, PDAs, tablet computers, laptops, all-in-one machines, smart driving devices, etc. It is understandable that the dialogue method provided by the embodiment of the present disclosure can also be applied in other scenarios.

[0045] like Figure 1 As shown, an embodiment of the present application provides a dialogue method, which may include the following steps.

[0046] Step 101: Receive a conversation request.

[0047] Step 102: Determine a conversation backtracking strategy based on the conversation request and stored conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information.

[0048] Step 103: Output the target reply content corresponding to the dialogue request according to the dialogue backtracking strategy.

[0049] In some feasible embodiments, a conversation request generally refers to the user's behavior of waking up the conversation model. The user can wake up by voice, by triggering a corresponding button, by inputting a corresponding text, or even by a corresponding gesture. If the user wakes up the large model by voice, the conversation request that triggers conversation backtracking can be a conversation request containing specific words. In some feasible embodiments, the specific word can be "continue", and the conversation request that triggers conversation backtracking can be a conversation request such as "continue chatting", "continue broadcasting", "continue the previous topic", "continue speaking" input by the user's voice. In some other feasible embodiments, the specific word can be "continue", and the conversation request that triggers conversation backtracking can be a conversation request such as "continue chatting", "continue broadcasting", "continue the previous topic", "continue speaking" input by the user's voice.

[0050] If the user wakes up the large model by pressing a button, then when the user needs to rewind the conversation, the user can trigger the button or icon on the interactive interface used to indicate conversation rewinding, or trigger the corresponding button on the device, so that the system will send a conversation request to trigger conversation rewinding to the large model.

[0051] In one feasible embodiment, the big model refers to the in-vehicle AI big model, which is a large artificial intelligence module based on deep learning technology and is designed for automotive intelligence. It consists of a huge neural network that can automatically extract and learn useful information from massive data, with high accuracy, efficiency and versatility. After training, these modules can process a variety of information such as language, images, and voice, and play a key role in fields such as autonomous driving and smart cockpits.

[0052] The key information of the conversation is the information stored when the user is having a conversation. In some feasible embodiments, the key information of the conversation only includes the information of one round of conversation, and one round of conversation includes a conversation question and the corresponding reply content of the conversation question. If multiple rounds of conversations are carried out during a conversation, until the conversation ends, the key information of the conversation of the first round of conversation will be stored in the first round of conversation, and the key information of the second round of conversation will be used to overwrite the key information of the conversation of the first round of conversation in the second round of conversation, and so on. In this way, when the conversation ends, only the key information of the conversation of the last round of conversation will be stored. When the user starts the second conversation, the key information of the conversation of the first round of conversation in the second conversation will overwrite the key information of the conversation of the last round of conversation in the first conversation. In this way, only a small amount of memory can be used to store the conversation content for backtracking for the user, reducing the system occupancy. In the above-mentioned in-vehicle AI big model, the key information of the conversation is stored every time the in-vehicle AI big model has a conversation. The in-vehicle AI big model conversation can be triggered by the user's direct voice call voice image trigger or the user's operation trigger on the in-vehicle host.

[0053] The key information of the conversation mainly includes the stored conversation content, the completeness of the conversation content, the conversation progress information, the information of the display card on the graphical user interface, and the additional content related to the conversation question. In some feasible embodiments, the conversation content includes the conversation question and the first reply content, wherein the conversation question may include: the question raised by the user, the classification of the question (car assistant, travel assistant, entertainment assistant, encyclopedia teacher, etc.), etc. The first reply content is the reply content of the previous round of conversation, and the first reply content may include: the text on the card (title serial number, etc.), video, picture, etc. The information of the display card on the graphical user interface may include: the type of card, the size of the card, etc. The additional content related to the conversation question may include: recommended resources (news, music, film and television, APP, etc.), related recommended questions, the source of information on the card, etc. The conversation progress information may include the output breakpoint of the text, the playback breakpoint of the video, etc.; the completeness of the conversation content may include a sign indicating that the reply content is generated and a sign indicating that the reply content is not generated, or the proportion of reply content generation, etc. This information is cached by the vehicle terminal to achieve a short-term memory effect. The key information of the conversation is stored in the memory in the form of variables. When using the key information, the corresponding memory address is found according to the variable name of the key information, and the value in the corresponding memory address is read, so that the key information of the conversation is obtained. The key information of the conversation stored in the memory will be cleared every time the conversation application (Android application package, APK) is restarted. This storage can reduce the access to the CPU (Central Processing Unit) and the hard disk, thereby improving the response speed. For example, if the conversation application is XX classmate set on the vehicle side, XX classmate will be closed when the vehicle is powered off and restarted when the vehicle is powered on. The key information of the conversation stored in the memory will be cleared every time the vehicle is powered on. If the user manually restarts XX classmate while the vehicle is continuously powered on, the key information of the conversation stored in the memory will also be cleared.

[0054] The conversation backtracking strategy refers to the strategy that needs to be executed when the conversation request includes at least one of the semantic keywords that continue, extend, supplement, and expand the previous conversation. For example, the received conversation request is "continue chatting", "add more", "expand some more", "is there anything else", etc.

[0055] In some feasible embodiments, in order to let the user know the function of conversation backtracking, the conversation method further includes: When receiving a signal that the conversation is interrupted, the interruption signal is synchronized to the voice image, and the voice image gives a retrospective prompt. Figure 2As shown, the user triggered the in-vehicle AI large model card during a power-on period, and before the broadcast was completed, the voice conflict scene interruption was triggered. When the in-vehicle AI large model business end receives the interruption signal of the large model card, it will synchronize the interruption signal to the voice image so that the voice image can prompt the user that there is an interruption signal. The interruption signal will be canceled after the broadcast is resumed or a new round of questions and answers begins. When any conversation is interrupted, an interruption signal will be generated. For example, if it is interrupted when executing the conversation backtracking strategy, the interruption signal will be synchronized, or if it is interrupted when a new conversation is in progress, the interruption signal will also be synchronized. When there is an interruption signal, the interruption signal will be synchronized to the voice image. The voice image can provide a backtracking prompt the next time the user wakes up the voice image, or it can provide a backtracking prompt the next time a conversation request is received, or it can provide a backtracking prompt immediately after receiving the interruption signal.

[0056] If the voice image gives a retrospective prompt the next time the user wakes up the voice image, or gives a retrospective prompt immediately after receiving an interruption signal, the user can choose to give up the conversation backtracking by operating the voice image. If the user chooses to give up the conversation backtracking through the voice image, the same interruption signal will not be prompted in the future. If the voice image gives a retrospective prompt the next time it receives a conversation request, at this time, if the user continues a new conversation, it means that the user has given up continuing to understand the content of the previous round of conversation. The key conversation information of the subsequent new round of question-and-answer conversation will overwrite the key conversation information stored when it was interrupted before, which will reduce the system memory usage. In order to avoid frequent prompts, each interruption signal will only be prompted once. Therefore, prompting through the voice image can prompt the user to recall that there was a conversation before the interruption, so that the user can go back to the conversation in time and get the content he wants.

[0057] The conversation backtracking strategy refers to the different response methods after the conversation backtracking is triggered. These response methods are determined based on the stored conversation key information. The target reply content refers to the content provided to the user after the conversation backtracking strategy is executed. It can be text, video, etc.

[0058] The conversation content includes the first reply content. After receiving the conversation request, a conversation backtracking strategy is determined according to the conversation request and the stored conversation key information, including: 1) Determine whether there is conversation key information to determine whether there is conversation content for backtracking. Conversation key information is the key to determining the conversation backtracking strategy. When the large model on the vehicle is triggered, the conversation key information will be stored. If there is no conversation key information stored, it means that the large model conversation has not been triggered before, or the stored conversation key information has been deleted when the application is restarted. At this time, there is no content to backtrack.

[0059] 2) According to the completeness of the conversation content, determine whether the first reply content has been generated. Since the large model generates answers in a streaming manner, it takes a certain amount of time for the large model to generate a complete answer. Since the timing of the large model conversation being interrupted and exited is random, the stored conversation key information will also change due to the content that the large model has generated. In some feasible embodiments, the conversation content includes the conversation question and the first reply content, and the first reply content can be text content, which is the content displayed on the card. Since the answer generated by the large module is streaming, when the completeness of the conversation content in the conversation key information indicates that the first reply content has been generated, the vehicle terminal has received a complete answer. If the completeness of the conversation content in the key information indicates that the first reply content has not been generated, the vehicle terminal has not received a complete answer. Specifically, whether the first reply content is complete can be determined based on the flag corresponding to the completeness of the conversation content, or whether the generation ratio of the first reply content reaches 100%. In some other embodiments, the first reply content can also be audio, video, etc. Regardless of the form of the first reply content, a completion mark can be added when the generation is completed through the large model, and the completion mark can be stored in the conversation key information. In this way, whether the first reply content is complete can be determined by judging whether there is a completion mark in the conversation key information.

[0060] 3) Determine whether the first reply content has been outputted according to the conversation progress information. In some feasible embodiments, the output may include voice broadcast and reply content display. Voice broadcast is to convert the text content into voice and broadcast it for the user to listen to. Reply content display is to display the reply content on a display screen or other display device for the user to view. Generally, the display of the reply content will be carried out as the large model is generated, and there is less chance of generating reply content display breakpoints. The speed of voice broadcast will be slower than the speed of reply content display, but the content of the voice broadcast corresponds to the text content. Therefore, the possibility of generating voice broadcast breakpoints is higher. In some feasible embodiments, the position of the voice broadcast breakpoint is used to determine whether the first reply content has been outputted. Generally, if the breakpoint of the voice broadcast is at the end of the text content, it can be determined that the output of the first reply content is completed, otherwise, the output of the first reply content is not completed.

[0061] 4) Determine a corresponding conversation backtracking strategy based on whether there is conversation content for backtracking, whether the first reply content has been generated, and whether the first reply content has been output, including: If there is no conversation content for backtracking, determining and executing a first conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated and the first reply content has been output, then determining and executing a second conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated but the first reply content has not been output, determining and executing a third conversation backtracking strategy; If there is conversation content for backtracking, and the first reply content has not been completely generated, a fourth conversation backtracking strategy is determined and executed.

[0062] According to the above technical means, the state of the previous round of conversation obtained by analyzing the key information of the conversation can be used as the basis for determining the conversation backtracking strategy, and the actual needs of the user to trigger the conversation backtracking are determined. Different conversation backtracking strategies are executed according to different needs to maximize the user's conversation backtracking needs and improve the user's conversation experience.

[0063] In some feasible embodiments, the first conversation backtracking strategy may include: starting a new conversation. There is no conversation content for backtracking, that is, there is no previous round of conversation, but the user has issued a conversation request, and the big model cannot but respond. Therefore, when there is no conversation content for backtracking, the big model can start a new conversation with the user to respond to the user's conversation needs, so that there must be a response to every request. In this way, the reply content generated by the new conversation is the target reply content corresponding to the conversation request in the first conversation backtracking strategy.

[0064] The second dialogue backtracking strategy may include: obtaining the dialogue question and the first reply content of the previous round from the dialogue key information; Synchronizing the conversation question and the first reply content in a contextual form; The reply supplementary content is generated according to the synchronized dialogue question and the first reply content, and the reply supplementary content is the target reply content corresponding to the dialogue request in the second dialogue backtracking strategy. In the second dialogue backtracking strategy, the target reply content may be partially the same as the first reply content, or it may be completely different.

[0065] There is conversation content for backtracking, that is, there is a previous round of conversation. After the conversation content is generated and the first reply content is output, it is determined that the user has consumed the content of the previous round of conversation. Therefore, the purpose of the user triggering conversation backtracking is the semantics of "continue generation", that is, the user wants to obtain more content based on the content that has been displayed. At this time, the big model supplements more content based on the previous round of answers, providing users with more required content and improving user experience.

[0066] The third conversation backtracking strategy may include: obtaining a first reply content and conversation progress information from the conversation key information, the first reply content being used as a target reply content in the third conversation backtracking strategy.

[0067] If there is conversation content for backtracking, and the first reply content has been generated, but the first reply content has not been output, in this case, the large model has generated complete reply content, but the user has not finished listening to the voice broadcast or has not finished reading the card content. At this time, the user's intention to trigger conversation backtracking should correspond to the semantics of "re-display and continue broadcasting", so it is necessary to obtain the first reply content and conversation progress information from the conversation key information, and prepare content for the subsequent strategy execution result output, so that the user can continue to read or listen to the answer that he read or listened to halfway in the previous round of conversation, thereby improving the user experience.

[0068] The fourth dialogue backtracking strategy may include: The previous round of dialogue questions are obtained from the dialogue key information, and new reply content is generated according to the dialogue question. The new reply content is the target reply content corresponding to the dialogue request in the fourth dialogue backtracking strategy. In the fourth dialogue backtracking strategy, the target reply content may be partially the same as the first reply content, or it may be completely different.

[0069] Since the answers generated by the big model are streaming, when the first reply content in the key information is incomplete, the car computer has not received the complete answer, and the first reply content has not been generated. At this time, the big model can only answer the same question in the previous round again, and then broadcast it normally. Although the consistency of the content generated by the two answers is not guaranteed at this time, it can provide complete answers to dialogue questions that the user has not received answers to before, thereby improving the user experience.

[0070] In some feasible embodiments, outputting target reply content corresponding to the dialogue request according to the dialogue backtracking strategy includes: Execute the dialogue backtracking strategy to obtain the target reply content; According to the target reply content, outputting the target reply content based on an output strategy; The output strategy includes a voice broadcast strategy and a reply content display strategy. The voice broadcast strategy mainly broadcasts the target reply content after the execution of the guide information and the dialogue backtracking strategy. The guide information is a configured dialogue template. Whenever the dialogue backtracking is performed, the same dialogue backtracking strategy will semantically broadcast the same dialogue template. The reply content after the dialogue backtracking strategy is executed is different according to different dialogue backtracking strategies.

[0071] If the first dialogue backtracking strategy is implemented in the conversation, the reply content of the new conversation needs to be voice broadcasted when the voice broadcast strategy is implemented. If the conversation question of the new conversation is what varieties of roses are there, the answer to the rose varieties will be broadcasted. If the second dialogue backtracking strategy is implemented in the conversation, the supplementary content of the reply needs to be voice broadcasted when the voice broadcast strategy is implemented. If the conversation question is what common diseases of roses are there, and powdery mildew and rust were answered in the previous round of conversation, and brown spot and black spot were answered in the supplementary content of the reply, then when the voice broadcast strategy is implemented, the answer including brown spot and black spot needs to be broadcasted; if the third dialogue backtracking strategy is implemented in the conversation, then when the voice broadcast strategy is implemented, the voice broadcast of the target reply content based on the conversation progress information needs to be continued. If the conversation question is what common diseases of roses are there, and powdery mildew and rust were answered in the previous round of conversation, and brown spot and black spot were answered in the supplementary content of the reply, then when the voice broadcast strategy is implemented, the answer including brown spot and black spot needs to be broadcasted. The reply content in the round of dialogue is powdery mildew and rust, but the dialogue output only outputs half of the reply content related to powdery mildew. When executing the voice broadcast strategy, it is necessary to continue to broadcast the other half of the reply content related to powdery mildew and the reply content related to rust according to the dialogue progress information; if the fourth dialogue backtracking strategy is executed in the dialogue, then when executing the voice broadcast strategy, it is necessary to voice broadcast the new reply content. If the dialogue question is what are the common diseases of roses, and the new reply content mainly introduces powdery mildew and brown spot, then when executing the voice broadcast strategy, it is necessary to broadcast the reply content introducing powdery mildew and brown spot.

[0072] The reply content display strategy mainly displays the execution results of the dialogue backtracking strategy, which can be only a text display of the target reply content, or a gesture image display of the target reply content. If the target reply content is a text display, it can be divided into a new dialogue display mode and a re-display mode according to whether the reply content in the dialogue backtracking strategy has been displayed to the user. The display mode and re-display mode of the new dialogue can be set according to the needs. For example, the display mode of the new dialogue can be a special style, such as gradually displaying with a typewriter effect. The typewriter effect can gradually display the reply content as the reply content is generated; the re-display mode can be without a special style, such as directly displaying all. Directly displaying all can facilitate users to directly view all reply content. If there is a voice broadcast, the user can quickly locate the output breakpoint position from the reply content in combination with the voice broadcast to avoid user reading delay caused by other effects. If there is no voice broadcast, the user can also quickly browse the content that has been read, and after browsing the content that has not been read before, read the unread content carefully.

[0073] In some feasible embodiments, the output strategy may include only a voice broadcast strategy, or only a reply content display strategy. The reply content display strategy may be divided into a text display strategy, a picture display strategy, a video display strategy, etc. according to different presentation forms. The presentation of the same content may be a single form, or a combination of two or more forms, for example, a single text display, a single picture display, a single video display, or a combination of text and picture, a combination of text and video, a combination of picture and video, or a combination of text, picture, and video. Therefore, the output strategy may also be a combination of a voice broadcast strategy and any reply content display strategy, for example: a combination strategy including a voice broadcast strategy and a text display strategy; and another example: a combination strategy including a voice broadcast strategy, a text display strategy, and a picture display strategy.

[0074] In some feasible embodiments, the voice broadcast strategy includes guide information output and target reply content output, and the guide information output and the target reply content output are linked by monitoring the guide information output end signal. After monitoring the guide information output end signal, the target reply content output is executed, which can avoid the target reply content output interrupting the guide information output and improve the integrity of the user's listening content. In a feasible embodiment, the voice broadcast end function (ttsend) can be called after the guide information output ends to send an end signal. After detecting that ttsend is called, it can be determined that the previous voice broadcast has ended.

[0075] The guidance information may include a new conversation start prompt voice when the first conversation backtracking strategy is executed, and an original conversation continuation prompt voice when the third conversation backtracking strategy or the fourth conversation backtracking strategy is executed. The guidance information is a configured conversation template, which can play a role in the conversation process and prompt the user the direction of the next conversation. For example, the new conversation start prompt voice can be "Okay, you come up with a topic.", "Okay, what do you want to talk about?", etc. After receiving such a response, the user will naturally say what he wants to talk about. After the big model obtains the content that the user wants to talk about, it generates a reply content based on these contents to start a new conversation. For another example, the original conversation continuation prompt voice can be "Okay, let's continue to talk about the previous topic.", etc. After receiving such a response, the user will know that he will hear the answer to the topic he asked before.

[0076] The target reply content output may include: output of the reply content of the new dialogue after executing the first dialogue backtracking strategy, and may specifically include voice broadcast of the reply content of the new dialogue.

[0077] Alternatively, the output of the reply supplementary content after executing the second dialogue backtracking strategy may specifically include voice broadcasting of the reply supplementary content.

[0078] Alternatively, after executing the third dialogue backtracking strategy, continuing to output the first reply content based on the dialogue progress information may specifically include continuing to voice broadcast the reply content based on the dialogue output breakpoint position. In some feasible embodiments, continuing to voice broadcast the reply content based on the dialogue output breakpoint position may first determine the sentence where the dialogue output breakpoint is located according to the dialogue output breakpoint position, and then voice broadcasting from the sentence where the output breakpoint is located.

[0079] Or the output of the new reply content after executing the fourth dialogue backtracking strategy may specifically include voice broadcasting of the new reply content.

[0080] The response content display strategy is mainly to display the execution results, including: After executing the first conversation backtracking strategy, the reply content of the new conversation is displayed according to the display mode of the new conversation. In a feasible embodiment, the display mode of the new conversation is to gradually display it using a typewriter effect.

[0081] After executing the second dialogue backtracking strategy, the reply supplementary content is displayed in the manner of displaying the new dialogue. In one feasible embodiment, the display manner of the new dialogue is to display it gradually using a typewriter effect.

[0082] Alternatively, after executing the third dialogue backtracking strategy, the first reply content is displayed in a re-display manner. In some feasible embodiments, the re-display manner is to directly display all reply content.

[0083] Or after executing the fourth dialogue backtracking strategy, the new reply content is displayed in the display mode of the new dialogue. In a feasible embodiment, the display mode of the new dialogue is to gradually display it using a typewriter effect.

[0084] In some feasible embodiments, the key conversation information is stored in the following manner: When the current round of conversation begins to generate reply content, the key conversation information of the previous round of conversation in the cache is cleared, and the key conversation information of the current round of conversation is stored in the cache. The current round of conversation can be a backtracking conversation or any other conversation that does not backtrack. The key conversation information of any conversation is cleared when the corresponding reply content is generated. If the conversation is backtracked, the conversation questions and reply content of the previous round of conversation in the key conversation information have been obtained at this time. Clearing the key conversation information will not affect the conversation backtracking, and it can also ensure that the key conversation information of the current round of conversation can be accurately stored.

[0085] like Figure 3As shown, after receiving a conversation request that triggers conversation backtracking, first determine whether there is conversation content for backtracking based on whether there is conversation key information in the storage. When there is no conversation content for backtracking, directly broadcast the new conversation start prompt voice to obtain the basis for reply content generation. In the case of conversation content for backtracking, further determine whether the first reply content is complete to determine whether the first reply content has been generated. If the first reply content has not been generated, restart the generation of the reply content and re-broadcast it. In this process, there are GPT generated labels and the expression of the voice broadcast content is inherited; if the first reply content has been generated, further determine whether the output of the first reply content has been completed. If the output of the first reply content has been completed, re-request semantics and continue to generate. In this process, there are GPT generated labels and the expression of the voice broadcast content is inherited; if the first reply content has not been output, re-expand the card and continue to output from the first reply content output breakpoint. In this process, there are GPT generated labels and the expression of the previous voice broadcast content is inherited.

[0086] The embodiment of the present application also provides another dialogue method, the method comprising: Step 201: receiving a dialogue request; Step 202: Determine a conversation backtracking strategy according to the conversation request and stored conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; Step 203: Output the target reply content corresponding to the dialogue request according to the dialogue backtracking strategy.

[0087] In the embodiment of the present application, the dialogue backtracking strategy includes: a first dialogue backtracking strategy, a second dialogue backtracking strategy, a third dialogue backtracking strategy and a fourth dialogue backtracking strategy. In the process of executing the dialogue backtracking strategy, it is necessary to store the dialogue key information. In some feasible embodiments, when the dialogue backtracking strategy is the first dialogue backtracking strategy, the dialogue key information of the new dialogue is stored in the cache. For example, when "continue chatting" is received and it is judged that there is no dialogue content for backtracking, the guiding information "OK, you come to start a topic" is broadcasted. Before the guiding information is broadcasted, the dialogue request "continue chatting" needs to be stored in the cache, and the guiding information of the answer "OK, you come to start a topic" is stored in the cache. When the dialogue question "What common diseases does rose have" is received, when the reply content is generated, the stored "continue chatting" and "OK, you come to start a topic" are cleared, and "What common diseases does rose have" is stored in the cache. Subsequently, according to the progress of the generation of the reply content, the stored dialogue key information is gradually improved.

[0088] When the conversation backtracking strategy is the second conversation backtracking strategy, when starting to generate the reply supplementary content, the conversation key information of the previous round of conversation in the cache is cleared, and the conversation request and the generated reply supplementary content are stored in the cache. For example, if the conversation question of the previous round of conversation is what common diseases do roses have, and the reply content of the previous round of conversation answers powdery mildew and rust, after receiving "continue chatting", it is determined according to the conversation backtracking strategy that a supplementary answer to "what common diseases do roses have" is needed, then the stored conversation question and reply content are cleared, and then "continue chatting" and the generated supplementary content are stored in the cache.

[0089] When the dialogue backtracking strategy is the third dialogue backtracking strategy, after obtaining the first reply content from the dialogue key information, the dialogue key information of the previous round of dialogue in the cache is cleared, the dialogue request is stored in the cache, and the first reply content is transferred to the cache as the target reply content. For example, if the dialogue question of the previous round of dialogue is what common diseases do roses have, and the reply content in the previous round of dialogue is powdery mildew and rust, but the dialogue output only outputs half of the reply content related to powdery mildew, after receiving "continue chatting", it is determined according to the dialogue backtracking strategy that it is necessary to continue answering "what common diseases do roses have", then after obtaining the reply content of the previous round, the dialogue question and reply content stored in the cache are cleared, and then "continue chatting" and the reply content of the previous round are transferred to the cache as the target reply content, that is, the reply content containing powdery mildew and rust generated in the previous round is stored.

[0090] When the conversation backtracking strategy is the fourth conversation backtracking strategy, when starting to generate new reply content, the conversation key information of the previous round of conversation in the cache is cleared, and the conversation request and the generated new reply content are stored in the cache. If the conversation question of the previous round of conversation is what common diseases do roses have, and the reply content in the previous round of conversation is powdery mildew and rust, but the reply content has not been generated, after receiving "continue chatting", it is determined according to the conversation backtracking strategy that an answer to "what common diseases do roses have" is required, then when starting to generate new reply content, the conversation key information of the previous round of conversation in the cache is cleared, and "continue chatting" and the generated new reply content are stored in the cache.

[0091] According to the above technical means, different conversation backtracking strategies correspond to different storage methods. When the conversation backtracking strategy needs to be generated, the key conversation information of the previous round of conversation in the cache is cleared when the generation begins, and the conversation request and the generated content are stored in the cache to ensure that the key conversation information is updated in real time. When the conversation backtracking strategy needs to backtrack and output the reply content of the previous round, the first reply content is transferred to the cache as the target reply content to ensure that the key conversation information can be stored normally, while also ensuring the accuracy of the key conversation information.

[0092] The embodiment of the present application also provides another dialogue method, which includes: Step 301: Receive a dialog request.

[0093] Step 302: Identify the intent of the conversation request. The intent of the conversation request refers to the purpose that the conversation request hopes to achieve. In this application, the intent of the conversation request refers to the function that the conversation request hopes the requested device to achieve. For example, when the requested device is a car computer, the intent of the conversation request can be the control of a functional unit on the car computer, or it can be a casual chat.

[0094] Step 303: Determine whether the intention is the intention of the target vertical domain. A vertical domain refers to a specific field divided by professional skills and business characteristics. Taking the car computer as an example, it can be divided into a chat vertical domain, an entertainment vertical domain, a control vertical domain, and so on. In the embodiment of the present application, the target vertical domain refers to the chat vertical domain, that is, the vertical domain for conversation.

[0095] Step 304: When the intention is the intention of the target vertical domain, determine whether the dialogue request triggers dialogue backtracking.

[0096] When the intention of the conversation request is to chat in the vertical domain, the conversation request can be the intention of directly conducting a conversation question and answer, or the intention of backtracking the previous round of conversation. For the intention of directly conducting a conversation question and answer, it is necessary to generate reply content based on the received conversation request and display it to the user. In some feasible embodiments, when the conversation request includes at least one of several semantic keywords that continue, extend, supplement, and expand the previous conversation, the conversation backtracking is triggered.

[0097] Step 305: when triggering the conversation backtracking, determining the conversation backtracking strategy according to the conversation request; Step 306: Output the target reply content corresponding to the dialogue request according to the dialogue backtracking strategy.

[0098] After receiving a conversation request, it can be determined whether to trigger conversation backtracking based on the recognized intent. When conversation backtracking is not triggered, a reply can be made directly. When conversation backtracking is triggered, a step of determining the conversation backtracking strategy is executed. This can avoid executing the step of determining the conversation backtracking strategy when the conversation request does not trigger conversation backtracking, thereby occupying system memory.

[0099] In some feasible embodiments, when triggering conversation backtracking, a conversation backtracking strategy may be determined according to the conversation request and the stored conversation key information, and the specific method will not be described in detail herein.

[0100] A second aspect of the present application provides a dialogue device, the dialogue device comprising: A data receiving unit, used for receiving a dialogue request; A strategy determination unit, configured to determine a conversation backtracking strategy according to the conversation request and stored conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; The execution unit is used to output the target reply content corresponding to the dialogue request according to the dialogue backtracking strategy.

[0101] According to the above-mentioned technical means, when the data receiving unit of the device receives a conversation request, the strategy determination unit determines the conversation backtracking strategy based on the stored conversation key information, and can perform conversation backtracking based on the stored conversation key information, providing the user with the possibility of obtaining the reply content again, thereby improving the user's conversation experience.

[0102] In some feasible embodiments, the conversation content includes a first reply content; the strategy determination unit includes a judgment module, and the judgment module is used to: Determine whether there is key information of the conversation to determine whether there is conversation content for backtracking; Determining whether the first reply content has been generated according to the completeness of the conversation content; Determine whether the first reply content has been outputted according to the conversation progress information.

[0103] In some feasible embodiments, the dialogue backtracking unit includes a strategy determination module, which is used to determine the dialogue backtracking strategy according to the judgment result of the judgment module, specifically: If there is no conversation content for backtracking, determining and executing a first conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated and the first reply content has been output, then determining and executing a second conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated but the first reply content has not been output, determining and executing a third conversation backtracking strategy; If there is conversation content for backtracking, and the first reply content has not been completely generated, a fourth conversation backtracking strategy is determined and executed.

[0104] In some possible embodiments, The first conversation backtracking strategy includes: starting a new conversation; The second dialogue backtracking strategy includes: obtaining the dialogue question and the first reply content of the previous round from the dialogue key information; Generate supplementary reply content according to the dialogue question and the first reply content; The third conversation backtracking strategy includes: obtaining the first reply content and conversation progress information from the conversation key information; The fourth dialogue backtracking strategy includes: The conversation questions of the previous round are obtained from the conversation key information, and new reply content is generated according to the conversation questions.

[0105] In some feasible embodiments, the execution unit further includes: a reply content generation module and an output display module; The reply content generation module is used to execute the dialogue backtracking strategy to obtain the target reply content; The output display module is used to output the target reply content based on the output strategy according to the target reply content; The output strategy includes a voice broadcast strategy and a reply content display strategy.

[0106] The output display module includes a voice broadcast module and an interface display module, and the output strategy includes a voice broadcast strategy and a reply content display strategy.

[0107] In some feasible embodiments, the voice broadcast strategy includes a guide information output and a target reply content output, and the guide information output and the target reply content output are linked by monitoring a guide information output end signal.

[0108] The guidance information includes a new conversation start prompt voice when the first conversation backtracking strategy is executed, and an original conversation continuation prompt voice when the third conversation backtracking strategy or the fourth conversation backtracking strategy is executed. The guidance information is a configured conversation template that can play a role in the conversation process and prompt the user the direction of the next conversation.

[0109] The target reply content output includes: output of the reply content of the new dialogue after executing the first dialogue backtracking strategy, which may specifically include a voice broadcast module performing voice broadcast on the reply content of the new dialogue.

[0110] Alternatively, the output of the reply supplementary content after executing the second dialogue backtracking strategy may specifically include a voice broadcast module performing voice broadcasting of the reply supplementary content.

[0111] Or after executing the third dialogue backtracking strategy, continuing to output the first reply content based on the dialogue progress information may specifically include a voice broadcast module continuing to voice broadcast the reply content based on the dialogue progress information.

[0112] Alternatively, the output of the new reply content after executing the fourth dialogue backtracking strategy may specifically include a voice broadcast module performing voice broadcasting on the new reply content.

[0113] The reply content display strategy is implemented through the interface display module, including: After executing the first conversation backtracking strategy, the interface display module displays the reply content of the new conversation in accordance with the display method of the new conversation.

[0114] After executing the second conversation backtracking strategy, the interface display module will display the reply supplementary content in the same way as a new conversation.

[0115] Or after executing the third conversation backtracking strategy, the interface display module displays the first reply content in a re-display manner.

[0116] Or after executing the fourth conversation backtracking strategy, the interface display module displays the new reply content in the display mode of the new conversation.

[0117] In some feasible embodiments, the execution unit further includes a storage unit, which is used to clear the key conversation information of the previous round of conversation in the cache and store the key conversation information of the current round of conversation in the cache when the current round of conversation starts to generate reply content.

[0118] In some feasible embodiments, the storage unit stores the conversation key information according to the execution of different conversation backtracking strategies, and when the conversation backtracking strategy is the first conversation backtracking strategy, the storage unit stores the conversation key information of the new conversation in the cache; When the conversation backtracking strategy is the second conversation backtracking strategy, when starting to generate the reply supplementary content, the storage unit clears the conversation key information of the previous round of conversation in the cache, and stores the conversation request and the generated reply supplementary content in the cache; When the conversation backtracking strategy is the third conversation backtracking strategy, after obtaining the first reply content from the conversation key information, the storage unit clears the conversation key information of the previous round of conversation in the cache, stores the conversation request in the cache, and transfers the first reply content as the target reply content to the cache; When the conversation backtracking strategy is the fourth conversation backtracking strategy, when starting to generate new reply content, the storage unit clears the conversation key information of the previous round of conversation in the cache, and stores the conversation request and the generated new reply content in the cache.

[0119] The third aspect of the present application provides a dialogue system, such as Figure 4 As shown, the system includes: a natural language understanding module, a chat application module, a large model module, a cache and an output module; The natural language understanding module NLU is used to understand the dialogue request and send the understood user intention to the chat application module; The chat application module ChatCopilot is used to query the stored conversation key information from the cache according to the user intention, and determine the conversation backtracking strategy according to the conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; The chat application module is further used to call the large model module and / or the natural language understanding module according to the dialogue backtracking strategy to determine the target reply content corresponding to the dialogue request; The output module is used to receive the target reply content sent by the chat application module and output the target reply content.

[0120] The natural language understanding module supports machine understanding of text content. In the application embodiment, the natural language understanding module realizes semantic recognition of received dialogue requests or dialogue questions. The chat application module is used to provide an interface for the vehicle end to call the large model module.

[0121] In some feasible embodiments, the system also includes an interaction module and a dialogue management module DM. During the dialogue process, the interaction module is used to realize the interaction between the natural language understanding module and the large model module and the dialogue management module. The dialogue management module is used to fully wake up the chat application module. The user intention understood by the natural language understanding module is first sent to the interaction module, and the interaction module then sends the user intention to the back-end dialogue management module, and the dialogue management module sends the user intention to the chat application module.

[0122] In some feasible embodiments, the conversation content includes the first reply content; the chat application module determines the conversation backtracking strategy in the following manner: Determine whether there is key information of the conversation to determine whether there is conversation content for backtracking; Determining whether the first reply content has been generated according to the completeness of the conversation content; Determining whether the first reply content has been outputted according to the conversation progress information; Determining a conversation backtracking strategy according to whether there is conversation content for backtracking, whether the first reply content has been generated, and whether the first reply content has been output; If there is no conversation content for backtracking, determining and executing a first conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated and the first reply content has been output, then determining and executing a second conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated but the first reply content has not been output, determining and executing a third conversation backtracking strategy; If there is conversation content for backtracking, and the first reply content has not been completely generated, a fourth conversation backtracking strategy is determined and executed.

[0123] In some feasible embodiments, the chat application module is further used to store the key conversation information in a cache after receiving the conversation intention identified by the natural language understanding module. The chat application module receives the conversation intention identified by the natural language understanding module, indicating that the reply content is about to be generated. If the conversation is backtracked, the conversation questions and reply content of the previous round of conversation in the key conversation information have been obtained at this time. Clearing the key conversation information will not affect the conversation backtracking, and it can also ensure that the key conversation information of the current round of conversation can be accurately stored. Specifically: After determining the conversation backtracking strategy, when the conversation backtracking strategy determined by the chat application module is the first conversation backtracking strategy, in the process of the chat application module calling the large model module, the natural language understanding module, and the output module to conduct a new conversation, the chat application module stores the conversation key information of the new conversation in the cache; When the conversation backtracking strategy determined by the chat application module is the second conversation backtracking strategy, when the large model module starts to generate the reply supplementary content, the chat application module clears the conversation key information of the previous round of conversation in the cache, and stores the conversation request and the generated reply supplementary content in the cache; When the conversation backtracking strategy determined by the chat application module is the third conversation backtracking strategy, after the chat application module obtains the first reply content from the cache, the chat application module clears the conversation key information of the previous round of conversation in the cache, stores the conversation request in the cache, and transfers the first reply content as the target reply content to the cache; When the conversation backtracking strategy determined by the chat application module is the fourth conversation backtracking strategy, when the large model module starts to generate new reply content, the chat application module clears the conversation key information of the previous round of conversation in the cache, and stores the conversation request and the generated new reply content in the cache.

[0124] According to the above technical means, different conversation backtracking strategies correspond to different storage methods. When the conversation backtracking strategy needs to be generated, the key conversation information of the previous round of conversation in the cache is cleared when the generation begins, and the conversation request and the generated content are stored in the cache to ensure that the key conversation information is updated in real time. When the conversation backtracking strategy needs to backtrack and output the reply content of the previous round, the first reply content is transferred to the cache as the target reply content to ensure that the key conversation information can be stored normally, while also ensuring the accuracy of the key conversation information.

[0125] In some feasible embodiments, the output module includes a voice broadcast module (Text To Speech, TTS) and a graphical user interface module (Graphical User Interface, GUI); If the conversation backtracking strategy determined by the chat application module is the first conversation backtracking strategy, the natural language understanding module identifies the semantics of the new conversation question and sends it to the large model module. The large model module generates new reply content according to the semantics of the new conversation question and sends it to the chat application module. The chat application module sends the new reply content to the graphical interface module for display and sends the new reply content to the voice broadcast module for broadcast. If the conversation backtracking strategy determined by the chat application module is the second conversation backtracking strategy, the chat application module requests the natural language understanding module to continue generating semantics, the natural language understanding module identifies the semantics of the previous round of conversation questions and sends it to the large model module, the large model module generates reply supplementary content according to the semantics of the previous round of conversation questions and sends it to the chat application module, the chat application module sends the reply supplementary content to the graphical interface module for display, and at the same time sends the reply supplementary content to the voice broadcast module for broadcasting; If the conversation backtracking strategy determined by the chat application module is the third conversation backtracking strategy, the chat application module obtains the first reply content as the target reply content, and sends it to the graphical interface module for display, and at the same time sends it to the voice broadcast module for voice broadcasting of the target reply content based on the conversation progress information; If the conversation backtracking strategy determined by the chat application module is the fourth conversation backtracking strategy, the chat application module requests the natural language understanding module for the previous round of conversation questions, the natural language understanding module identifies the semantics of the previous round of conversation questions and sends them to the large model module, the large model module generates new reply content according to the semantics of the previous round of conversation questions and sends it to the chat application module, the chat application module sends the new reply content to the graphical interface module for display, and at the same time sends the new reply content to the voice broadcast module for broadcast.

[0126] According to the above technical means, different reply contents are generated according to different conversation backtracking strategies and displayed and outputted. According to the overall situation of the user's previous round of conversation, the reply contents can be outputted in a targeted manner to improve the user's conversation experience.

[0127] In some feasible embodiments, the system further comprises a natural language generation module NLG. The guidance information is broadcasted by the natural language generation module, and the guidance information broadcast is linked to the target reply content broadcast by monitoring a guidance information broadcast end signal.

[0128] The guidance information broadcast includes a new dialogue start prompt voice broadcast when the natural language generation module executes the first dialogue backtracking strategy, and an original dialogue continuation prompt voice broadcast when the third dialogue backtracking strategy or the fourth dialogue backtracking strategy is executed.

[0129] In some feasible embodiments, the natural language understanding module and the large model module are both deployed in the cloud, and the natural language understanding module and the large model module interact with the chat application module through the interaction module and the dialogue management module.

[0130] The following takes the reply content as text content as an example, and combines the dialogue system to further illustrate the specific interaction process of the dialogue method provided by this application.

[0131] When there is no dialogue content for backtracking, determine and execute the first dialogue backtracking strategy. Figure 5 The first backtracking strategy interaction diagram of this application is shown as follows: Figure 5 As shown in the figure, the user triggers the natural language understanding module (NLU) to recognize the conversation backtracking intention of the voice through the voice containing the trigger word "continue" and issues a search content instruction to send the conversation backtracking intention to the interaction module. The interaction module wakes up the dialogue management module (DM) and sends the conversation backtracking intention to the dialogue management module. The dialogue management module fully wakes up the chat application module (ChatCopilot) and sends the conversation backtracking intention. The chat application module receives the conversation backtracking intention and starts to query whether there is backtracking content in the cache and determines whether conversation backtracking is supported. When there is no backtracking content in the cache, the chat application module returns a successful code (successful_no_content) to the dialogue management module, and the dialogue management module returns a successful code (successful_no_content) to the natural language generation module (NLG). The natural language generation module broadcasts the new conversation start prompt voice "Okay, you come up with a topic." When there is no backtracking content, it can be determined that there is no previous round of conversation. At this time, the new conversation start prompt voice is played to obtain the content that the user wants to talk to, so that the conversation can be carried out according to the user's needs and the user experience can be improved.

[0132] If there is conversation content for backtracking, and the first reply content has been generated and output, that is, there is backtracking content, the user has seen all the content and has finished listening to the voice broadcast, in this case, it is believed that the user has consumed the current question and answer, so the "continue chatting" intention should correspond to the "continue generating" semantics, so we make a second request for the semantics, and this request must send the "continue generating" semantics, then determine and execute the second conversation backtracking strategy.

[0133] like Figure 6The second backtracking strategy interaction diagram of this application is shown as follows: Figure 6 As shown, the user triggers the natural language understanding module to recognize the conversation backtracking intention of the voice through the voice "continue" containing the trigger word "continue" and issues a content search instruction to send the conversation backtracking intention to the interaction module. The interaction module wakes up the dialogue management module and sends the dialogue backtracking intention to the dialogue management module. The dialogue management module fully wakes up the chat application module and sends the dialogue backtracking intention. The chat application module receives the dialogue backtracking intention and starts to query whether there is backtracking content in the cache and determines whether dialogue backtracking is supported. When there is backtracking content in the cache, content generation is completed, and voice broadcasting is completed, the chat application module returns a successful code (successful_tts_finished) to the dialogue management module. At the same time, the chat application module sends a request to continue generating semantics to the natural language understanding module, and the request to continue generating semantics is synchronized to the natural language understanding module via the interaction module. The natural language understanding module identifies the dialogue questions and issues a search content instruction to synchronize the understanding results of the previous round of dialogue questions and the first reply content to the chat application module via the interaction module and the dialogue management module, and send them to the large model module at the same time. The large model module generates reply supplementary content based on the understanding results and sends it to the chat application module via the interaction module and the dialogue management module. The chat application module provides the reply supplementary content to the graphical interface module (GUI) to gradually display the cards using a typewriter effect, and at the same time provides it to the voice broadcast module, which broadcasts the reply supplementary content.

[0134] If there is conversation content for backtracking, and the first reply content has been generated but not output, in this case, the terminal has received the complete answer, but the user has not finished listening to the voice broadcast or reading the card content. At this time, the "continue chatting" intention should correspond to the semantics of "re-display and continue broadcasting", then determine and execute the third conversation backtracking strategy.

[0135] like Figure 7 The third backtracking strategy interaction diagram of this application is shown as follows: Figure 7As shown, the user triggers the natural language understanding module to recognize the conversation backtracking intention of the voice through the voice containing the trigger word "continue" and sends a search content instruction to send the conversation backtracking intention to the interaction module. The interaction module wakes up the dialogue management module and sends the conversation backtracking intention to the dialogue management module. The dialogue management module fully wakes up the chat application module and sends the conversation backtracking intention. The chat application module receives the conversation backtracking intention and starts to query whether there is backtracking content in the cache and judge whether the conversation backtracking is supported. If there is backtracking content in the cache and the content generation is completed, but the voice broadcast is not completed, the chat application module returns a successful code (successful) to the dialogue management module, and the dialogue management module returns a successful code (successful) to the natural language generation module. The natural language generation module broadcasts the original conversation to continue the prompt voice. For example, after the broadcast of "OK, let's continue to talk about the previous topic" is completed, the dialogue management module is notified, and the dialogue management module notifies the chat application module. The chat application module provides the reply content of the previous round to the graphical interface module to directly display the card. At the same time, the chat application module provides the reply content of the previous round and the dialogue output breakpoint position to the voice broadcast module, and the voice broadcast module broadcasts the content after the interruption.

[0136] If there is dialogue content for backtracking and the first reply content has not been generated, in this case, because the answer generated by the large model module is streaming, the terminal side has not received the complete answer, so the large model module can only answer the same question in the previous round again. At this time, the fourth dialogue backtracking strategy is determined and executed.

[0137] like Figure 8 The fourth backtracking strategy interaction diagram of this application is shown as follows: Figure 8As shown, the user triggers the natural language understanding module on the vehicle side to recognize the conversation backtracking intention of the voice through the voice containing the trigger word "continue" and issue a search content instruction to send the conversation backtracking intention to the interaction module. The interaction module wakes up the dialogue management module and sends the dialogue backtracking intention to the dialogue management module. The dialogue management module fully wakes up the chat application module and sends the dialogue backtracking intention. The chat application module receives the dialogue backtracking intention and starts to query whether there is backtracking content in the cache and determines whether dialogue backtracking is supported. When there is backtracking content in the cache but the content has not been generated and the voice broadcast is not completed, the chat application module returns a success code (successful) to the dialogue management module, and the dialogue management module returns a success code (successful) to the natural language generation module. The natural language generation module broadcasts the original dialogue continuation prompt voice. For example, after the broadcast of "OK, let's continue to talk about the previous topic" is completed, the dialogue management module is notified, and the dialogue management module then notifies the chat application module. The chat application module requests the dialogue question (query) of the previous round of questions and answers and synchronizes it to the natural language understanding module through the interaction module. The natural language understanding module identifies the dialogue question and issues a search instruction content. The understanding result is synchronized to the chat application module through the interaction module and the dialogue management module, and sent to the large model module at the same time. The large model module generates a reply content based on the understanding result and sends it to the chat application module through the interaction module and the dialogue management module. The chat application module provides the new reply content to the graphical interface module (GUI) to gradually display the card using a typewriter effect, and provides it to the voice broadcast module at the same time. The voice broadcast module broadcasts the new reply content. Therefore, when the backtracking dialogue content is incomplete, new reply content is generated according to the dialogue question, which can provide complete answers to dialogue questions that users have not received answers to before, thereby improving user experience.

[0138] In some feasible embodiments, during the voice broadcast process, the expression of the current voice broadcast content is set to indicate which category the current question and answer service belongs to. For example, for entertainment question and answer services, cheerful and relaxed expressions can be set, such as the expression of listening to music, etc.; for encyclopedia question and answer services, rigorous and formal expressions can be set, such as the expression of wearing teaching aids, the expression of wearing glasses, etc. The conversation backtracking also sets expressions to indicate the business type of the current conversation, and the specific expressions can be set according to user needs.

[0139] Conversation backtracking is equivalent to a normal round of conversation interaction, so it is also necessary to synchronize context information to the natural language understanding module for semantic arbitration. During the conversation backtracking process, the business notification interaction module caches and applies the context.

[0140] Each time a large-module non-dialogue backtracking intent is received, the notification interaction module needs to cache the reply content of this round; each time a dialogue backtracking intent is received, the notification interaction module needs to add the cached content of the previous round to the context list.

[0141] A fourth aspect of the present application provides an electronic device, including: a memory configured to store instructions; and A processor is configured to call the instructions from the memory and implement the dialog method when executing the instructions.

[0142] A fifth aspect of the present application provides a vehicle, including a control system, which applies the dialogue method.

[0143] A sixth aspect of the present application provides a machine-readable storage medium, on which instructions are stored, and the instructions are used to enable a machine to execute the described dialogue method.

[0144] A seventh aspect of the present application provides a computer program product, comprising a computer program, which implements the dialogue method when executed by a processor.

[0145] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0146] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0147] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0148] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0149] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0150] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0151] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0152] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0153] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A conversation method, characterized in that: The dialogue method comprises: Receive a conversation request; Determine a conversation backtracking strategy according to the conversation request and the stored conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; According to the dialogue backtracking strategy, the target reply content corresponding to the dialogue request is output.

2. The dialogue method according to claim 1, characterized in that: The conversation content includes the first reply content; Determining a dialogue backtracking strategy according to the dialogue request and the stored dialogue key information includes: Determine whether there is key information of the conversation to determine whether there is conversation content for backtracking; Determining whether the first reply content has been generated according to the completeness of the conversation content; Determining whether the first reply content has been outputted according to the conversation progress information; A conversation backtracking strategy is determined according to whether there is conversation content for backtracking, whether the first reply content has been generated, and whether the first reply content has been output.

3. The dialogue method according to claim 2, characterized in that: Determining a conversation backtracking strategy according to whether there is conversation content for backtracking, whether the first reply content has been generated, and whether the first reply content has been output, includes: If there is no conversation content for backtracking, determining and executing a first conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated and the first reply content has been output, then determining and executing a second conversation backtracking strategy; If there is a conversation content for backtracking, and the first reply content has been generated but the first reply content has not been output, determining and executing a third conversation backtracking strategy; If there is conversation content for backtracking, and the first reply content has not been completely generated, a fourth conversation backtracking strategy is determined and executed.

4. The dialogue method according to claim 1, characterized in that: The key information of the conversation is stored in the following way: When the reply content of the current round of dialogue begins to be generated, the key dialogue information of the previous round of dialogue in the cache is cleared, and the key dialogue information of the current round of dialogue is stored in the cache.

5. The dialogue method according to claim 1, characterized in that: The dialogue backtracking strategy includes: a first dialogue backtracking strategy, a second dialogue backtracking strategy, a third dialogue backtracking strategy and a fourth dialogue backtracking strategy; The first conversation backtracking strategy includes: starting a new conversation; The second dialogue backtracking strategy includes: obtaining the dialogue question and the first reply content of the previous round from the dialogue key information; Generate supplementary reply content according to the dialogue question and the first reply content; The third conversation backtracking strategy includes: obtaining the first reply content and conversation progress information from the conversation key information; The fourth dialogue backtracking strategy includes: The conversation questions of the previous round are obtained from the conversation key information, and new reply content is generated according to the conversation questions.

6. The dialogue method according to claim 1, characterized in that: The dialogue backtracking strategy includes: a first dialogue backtracking strategy, a second dialogue backtracking strategy, a third dialogue backtracking strategy and a fourth dialogue backtracking strategy; After determining the conversation backtracking strategy, when the conversation backtracking strategy is the first conversation backtracking strategy, storing the conversation key information of the new conversation in the cache; When the conversation backtracking strategy is the second conversation backtracking strategy, when starting to generate the reply supplementary content, clear the conversation key information of the previous round of conversation in the cache, and store the conversation request and the generated reply supplementary content in the cache; When the conversation backtracking strategy is the third conversation backtracking strategy, after obtaining the first reply content from the conversation key information, clearing the conversation key information of the previous round of conversation in the cache, storing the conversation request in the cache, and transferring the first reply content as the target reply content to the cache; When the conversation backtracking strategy is the fourth conversation backtracking strategy, when starting to generate new reply content, the conversation key information of the previous round of conversation in the cache is cleared, and the conversation request and the generated new reply content are stored in the cache.

7. The dialogue method according to claim 5, characterized in that: Outputting target reply content corresponding to the dialogue request according to the dialogue backtracking strategy includes: Execute the dialogue backtracking strategy to obtain the target reply content; According to the target reply content, outputting the target reply content based on an output strategy; The output strategy includes a voice broadcast strategy and a reply content display strategy.

8. The dialogue method according to claim 7, characterized in that: The voice broadcast strategy includes guide information output and target reply content output, and the guide information output and the target reply content output are linked by monitoring the guide information output end signal; The guidance information includes a new dialogue start prompt voice when the first dialogue backtracking strategy is executed, and an original dialogue continuation prompt voice when the third dialogue backtracking strategy or the fourth dialogue backtracking strategy is executed; The target reply content output includes: outputting the reply content of the new dialogue after executing the first dialogue backtracking strategy; or outputting the supplementary content of the reply after executing the second dialogue backtracking strategy; or after executing the third conversation backtracking strategy, continuing to output the first reply content based on the conversation progress information; Or the output of the new reply content after executing the fourth dialogue backtracking strategy.

9. The dialogue method according to claim 1, characterized in that: After receiving the dialog request and before determining the dialog backtracking strategy, the method further includes: identifying the intent of the conversation request; Determine whether the intention is the intention of the target vertical domain; When the intention is the intention of the target vertical domain, determining whether the dialog request triggers dialog backtracking; When the dialog backtracking is triggered, the dialog backtracking strategy determination step is executed.

10. The dialogue method according to claim 1, characterized in that: The dialogue method further includes: When a conversation interruption signal is received, the interruption signal is synchronized to the voice image, and the voice image performs a backtracking prompt.

11. A conversation device, characterized in that: The dialogue device comprises: A data receiving unit, used for receiving a dialogue request; A strategy determination unit, configured to determine a conversation backtracking strategy according to the conversation request and stored conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; The execution unit is used to output the target reply content corresponding to the dialogue request according to the dialogue backtracking strategy.

12. A dialogue system, characterized in that: The system includes: a natural language understanding module, a chat application module, a large model module, a cache and an output module; The natural language understanding module is used to understand the dialogue request and send the understood user intention to the chat application module; The chat application module is used to query the stored conversation key information from the cache according to the user intention, and determine the conversation backtracking strategy according to the conversation key information; the conversation key information includes: conversation content, conversation content completeness and conversation progress information; The chat application module is further used to call the large model module and / or the natural language understanding module according to the dialogue backtracking strategy to determine the target reply content corresponding to the dialogue request; The output module is used to receive the target reply content sent by the chat application module and output the target reply content.

13. The dialogue system according to claim 12, characterized in that: The output module includes a voice broadcast module and a graphic interface module; If the conversation backtracking strategy determined by the chat application module is the second conversation backtracking strategy, the chat application module requests the natural language understanding module to continue generating semantics, the natural language understanding module identifies the semantics of the previous round of conversation questions and sends it to the large model module, the large model module generates reply supplementary content according to the semantics of the previous round of conversation questions and sends it to the chat application module, the chat application module sends the reply supplementary content to the graphical interface module for display, and at the same time sends the reply supplementary content to the voice broadcast module for broadcasting; If the conversation backtracking strategy determined by the chat application module is the third conversation backtracking strategy, the chat application module obtains the first reply content as the target reply content, and sends it to the graphical interface module for display, and at the same time sends it to the voice broadcast module for voice broadcasting of the target reply content based on the conversation progress information; If the conversation backtracking strategy determined by the chat application module is the fourth conversation backtracking strategy, the chat application module requests the natural language understanding module for the previous round of conversation questions, the natural language understanding module identifies the semantics of the previous round of conversation questions and sends them to the large model module, the large model module generates new reply content according to the semantics of the previous round of conversation questions and sends it to the chat application module, the chat application module sends the new reply content to the graphical interface module for display, and at the same time sends the new reply content to the voice broadcast module for broadcast.

14. An electronic device, characterized in that: include: a memory configured to store instructions; as well as A processor is configured to call the instructions from the memory and implement the dialog method according to any one of claims 1 to 10 when executing the instructions.

15. A vehicle, comprising a control system, characterized in that: The control system applies the dialog method as claimed in any one of claims 1 to 10.

16. A machine-readable storage medium, characterized in that: The machine-readable storage medium stores instructions for causing a machine to execute the dialogue method according to any one of claims 1 to 10.

17. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the dialog method according to any one of claims 1 to 10 is implemented.

Images