Intelligent cabin dialogue processing method and device, equipment and medium
By performing preliminary processing and intent analysis on the dialogue information between the user and the in-vehicle system, the system generates response text or function parameters by calling the large language model only once, which solves the problem of time-consuming dialogue processing in existing technologies and achieves rapid response and intelligent interaction.
Patent Information
- Application Number
- CN202511333397.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-18
- Publication Date
- 2025-12-16
AI Technical Summary
In existing intelligent cockpit dialogue processing methods, the model is invoked serially, which results in significant time consumption and is prone to end-to-end response timeout issues.
By performing preliminary processing and intent analysis on the current dialogue information between the user and the vehicle system, the target intelligent agent label is determined. The current dialogue information and functional materials are then input into the large language model. The large language model is called only once to generate the response text or functional parameters, and the corresponding vehicle system function is called.
It improves end-to-end response speed, enables faster response to user requests, reduces operational complexity and distraction, and enhances driving safety and system intelligence.
Smart Images
Figure CN121145879A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of vehicle technology, and in particular to a method, apparatus, device and medium for processing intelligent cockpit dialogue. Background Technology
[0002] In recent years, with the rapid development of artificial intelligence (AI) technology, natural language processing and automatic speech recognition technologies have been increasingly widely used in intelligent dialogue systems. In scenarios such as enterprise management, customer service, and smart cockpits, the demand for users to interact with systems via voice is constantly growing.
[0003] In existing technologies, the user's voice input query question is converted into natural language description text and corresponding unanswered questions. The natural language description text is then used to generate SQL statements. The natural language description text, SQL statements, etc., are combined into a tuple describing the retrieval link, which is then input into a large language model. The test results are output, and the retrieval results with qualified test results and corresponding unanswered questions are then input into the large language model. Combined with retrieval enhancement generation technology, an answer is generated and returned to the user.
[0004] However, the above dialogue processing flow calls the model serially and uses the large language model for retrieval multiple times, resulting in significant time consumption and a high risk of end-to-end response timeouts. Summary of the Invention
[0005] In view of the above problems, this application proposes a method, apparatus, device, and medium for processing intelligent cockpit dialogue to solve these problems. It first performs preliminary processing and intent analysis on the current dialogue information between the user and the vehicle system to obtain functional material related to the user's intent. Then, the current dialogue information and functional material are input into a large language model to obtain the response text output by the large language model or the functional parameters of the target functional function. Furthermore, it can invoke the corresponding vehicle system function to achieve intelligent interaction between the user and the vehicle system. This method only requires calling the large language model once, improving end-to-end response speed and enabling faster responses to user requests.
[0006] Firstly, this application provides a method for processing dialogue in an intelligent cockpit, the method comprising: Obtain current dialogue information, which includes the current request text and the current dialogue scenario value; Based on the current dialogue information, the target intelligent agent label is determined from a set of preset intelligent agent labels; Based on the current dialogue information and the target agent's label, retrieve one or more preliminary function selections from a preset function library; The initially selected function will be used as the function source material. The current dialogue information and the functional material are input into a pre-built large language model to obtain the response text or functional parameters of the target functional function output by the large language model. If the output of the large language model is the function parameter, then the vehicle system function corresponding to the target function is called, and the parameters of the vehicle system function are adjusted according to the function parameter.
[0007] Optionally, determining the target agent tag from a set of preset agent tags based on the current dialogue information includes: The vehicle's current status, the available resources of the in-vehicle system, and historical dialogue information are obtained. The current status includes the status of the in-vehicle equipment and the driving status. Based on the current dialogue information, the current state, the available resources, and the historical dialogue information, the target agent label is determined from a set of preset agent labels.
[0008] Optionally, determining the target agent tag from a set of preset agent tags based on the current dialogue information, the current state, the available resources, and the historical dialogue information includes: The target keyword corresponding to the current dialogue information is matched from the preset keyword template library, and the first agent label and the first matching degree corresponding to the target keyword are determined. The first matching degree is the matching degree between the current dialogue information and the first agent label. Based on the current dialogue information, the current state, the available resources, and the historical dialogue information, a second agent label and a second matching degree are determined from a set of preset agent labels. The second matching degree is the matching degree between the current dialogue information, the current state, the available resources, and the historical dialogue information and the second agent label. If the first matching degree is greater than the second matching degree, then the first agent label is used as the target agent label; If the first matching degree is less than or equal to the second matching degree, then the second agent label is used as the target agent label.
[0009] Optionally, the keyword template library includes a first keyword template library and a second keyword template library, and the step of matching the target keyword corresponding to the current dialogue information from the preset keyword template library includes: A template matching algorithm is used to match the first matching keyword corresponding to the current dialogue information from the first keyword template library, and to determine the first similarity between the current dialogue information and the first matching keyword. A fuzzy matching algorithm is used to match the second matching keyword corresponding to the current dialogue information from the second keyword template library, and to determine the second similarity between the current dialogue information and the second matching keyword. If the first similarity is greater than the second similarity, then the first matching keyword is taken as the target keyword; If the first similarity is greater than or equal to the second similarity, then the second matching keyword is used as the target keyword.
[0010] Optionally, the method further includes: The larger of the first similarity and the second similarity is taken as the target similarity; If the target similarity is greater than the preset similarity threshold, then the matching function corresponding to the target keyword will be used as the function material.
[0011] Optionally, the step of retrieving one or more preliminary function selections from a preset function library based on the current dialogue information and the target agent label includes: Determine the correlation between the current dialogue information and the target agent label and each function in the preset function library; Functions with a relevance greater than a preset relevance threshold are selected as the initial function functions.
[0012] Optionally, determining the relevance of the current dialogue information and the target agent label to each function in a preset function library includes: The current dialogue information and the target agent label are converted into retrieval vector data; Calculate the vector similarity between the retrieved vector data and the function vector data corresponding to each function in the preset function library; Based on vector similarity, the correlation between the current dialogue information and the target agent label and each functional function is determined.
[0013] Secondly, this application provides a processing device for intelligent cockpit dialogue, the device comprising: The acquisition module is used to acquire current dialogue information, which includes the current request text and the current dialogue scenario value; The first determining module is used to determine the target intelligent agent label from a plurality of preset intelligent agent labels based on the current dialogue information; The retrieval module is used to retrieve one or more initially selected function functions from a preset function function library based on the current dialogue information and the target intelligent agent label; The material module is used to use the initially selected function as function material; The second determining module is used to input the current dialogue information and the functional material into a pre-constructed large language model to obtain the response text or functional parameters of the target functional function output by the large language model. The control module is used to call the vehicle system function corresponding to the target function function if the output of the large language model is the function parameter, and adjust the parameters of the vehicle system function according to the function parameter.
[0014] Thirdly, this application provides an electronic device, including: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method as described in the first aspect.
[0015] Fourthly, this application provides a computer-readable storage medium storing computer instructions for causing the computer to perform the method described in the first aspect.
[0016] The technical solutions provided in this application embodiment have at least the following technical effects or advantages: This application provides a method, apparatus, device, and medium for processing intelligent cockpit dialogue. The method involves acquiring current dialogue information, obtaining user input commands and dialogue scenarios, including the current request text and the current dialogue scenario value. Based on the current dialogue information, a target intelligent agent label is determined from a set of preset intelligent agent labels to initially analyze the user's intent. Based on the current dialogue information and the target intelligent agent label, one or more preliminary function selections are retrieved from a preset function library to further analyze ways to resolve the user's intent. These preliminary function selections are used as function function materials to obtain several possible user intents. The current dialogue information and function function materials are input into a pre-constructed large language model to obtain the response text output by the large language model or the function parameters of the target function. The method analyzes the most likely user intent and responds accordingly. If the large language model outputs function parameters, the corresponding in-vehicle system function is called, and the parameters of the in-vehicle system function are adjusted according to the function parameters. If the user wants to control the in-vehicle system function, the corresponding in-vehicle system function is executed. This method can first perform preliminary processing and intent analysis on the current dialogue information between the user and the vehicle system to obtain functional materials related to the user's intent. Then, the current dialogue information and functional materials are input into a large language model to obtain the response text or functional parameters of the target functional function output by the large language model. It can also call the vehicle system function corresponding to the target functional function to realize intelligent interaction between the user and the vehicle system. Only one call to the large language model is needed, which improves the end-to-end response speed and can respond to user requests more quickly.
[0017] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description
[0018] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart of a smart cockpit dialogue processing method provided in an embodiment of this application; Figure 2 This is a structural block diagram of a smart cockpit dialogue processing device provided in an embodiment of this application. Detailed Implementation
[0019] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the embodiments of this disclosure and the specific features in the embodiments are detailed descriptions of the technical solutions of this application, rather than limitations on the technical solutions of this application. Unless otherwise specified, the embodiments of this application and the technical features in the embodiments can be combined with each other.
[0020] Figure 1 This is a flowchart of a smart cockpit dialogue processing method provided in an embodiment of this application, such as... Figure 1 As shown, the method includes: Step S110: Obtain the current dialogue information, which includes the current request text and the current dialogue scenario value.
[0021] In this embodiment, the user can input their request to the vehicle system via voice, text or image. If the input is voice or image, the voice or image is converted into text by voice recognition technology or image recognition technology to form the user's request text, which serves as the request instruction.
[0022] The scenario value is used to characterize the specific domain or task category to which the user's input request text belongs, essentially acting as a key label for primary classification of the user's intent. For example, the user's intent may fall into categories such as vehicle control, navigation and travel, entertainment and information, communication and social interaction, or casual conversation and question answering; different categories correspond to different scenario values.
[0023] Step S120: Based on the current dialogue information, determine the target agent label from a set of preset agent labels.
[0024] In this embodiment, the in-vehicle system includes multiple intelligent agents, such as a vehicle control agent, a navigation and travel agent, an entertainment and information agent, a communication and social agent, and a chat and question-answering agent. Each agent has a corresponding agent tag, and the corresponding agent is invoked through the agent tag to infer user intent. Therefore, based on the current dialogue information, the user's intent can be preliminarily analyzed, and the agent that can resolve the user's intent can be identified as the target agent. Simultaneously, the target agent tag corresponding to the target agent can be found.
[0025] Step S130: Based on the current dialogue information and the target agent's label, retrieve one or more preliminary function selections from the preset function library.
[0026] In this embodiment, a recall module can be built in advance, which stores a function library. The recall module can be built based on an offline MongoDB database and an online Elasticsearch (ES) engine. MongoDB, as a document-oriented database, stores structured and semi-structured vehicle function data, such as function definitions, vehicle function descriptions and operation guides, and frequently asked questions; while ES has powerful full-text search capabilities. The function library stores multiple functions, such as vehicle control functions (air conditioning, windows, and lights), entertainment and information functions (music and video), and navigation and travel functions (navigation and query functions).
[0027] The current dialogue information and target agent tags are input into the recall module. The recall module uses the retrieval features of Elasticsearch to retrieve one or more initial function selections in a short time without generating additional SQL statements, thus improving retrieval efficiency.
[0028] Step S140: Use the initially selected function as the function material.
[0029] In this embodiment of the application, a function material is constructed using a preliminary selection of function functions. That is, the function material includes one or more preliminary selection function functions. The preliminary selection function functions are function functions related to the user intent that has been initially analyzed, and are used for further retrieval in the future.
[0030] Step S150: Input the current dialogue information and function material into the pre-built large language model to obtain the response text or function parameters of the target function output by the large language model.
[0031] In this embodiment, a large language model can be constructed based on the ReAct (Reason + Act) architecture and a large amount of in-vehicle scenario data, including but not limited to driving preference data, vehicle function data, and user interaction data. This large model is then optimized and trained to deepen the understanding of various semantic information and user needs in in-vehicle scenarios. The specific construction method is existing technology and will not be elaborated here.
[0032] Specifically, the large language model determines the target function from the function material based on the current dialogue information input; if the target function is a question-and-answer type, it generates a response text to the user's request based on the definition of the target function and the answer resource generated based on the current request text, that is, it provides an answer to the user's question; if the target function is a control type, it generates the function parameters of the target function.
[0033] For example, if a user says "What is the temperature today?" via voice, the identified agent label is the question-and-answer agent label, the target function generated by the large language model is the question-and-answer encyclopedia function, and the reply text is "Today is 20℃".
[0034] Step S160: If the output of the large language model is a function parameter, then call the vehicle system function corresponding to the target function function, and adjust the parameters of the vehicle system function according to the function parameter.
[0035] In this embodiment of the application, if the output of the large language model is a function parameter, it means that the user wants to control the control function of the vehicle. Then, the vehicle system function corresponding to the target function function is called, and the parameters of the vehicle system function are adjusted according to the function parameter to meet the user's needs.
[0036] For example, if a user says "It's too hot in the car" via voice, the target intelligent agent label is determined as the vehicle control intelligent agent label, the target function generated by the large language model is the air conditioning function, and the air conditioning parameters of the air conditioning function are temperature 24℃ and fan speed medium. Then, the air conditioning is turned on and the temperature is set to 24℃ and the fan speed is set to medium.
[0037] Optionally, step S120 includes: Step S1201: Obtain the current status of the vehicle, the available resources of the vehicle system, and historical dialogue information. The current status includes the status of the vehicle equipment and the driving status.
[0038] In this embodiment, the in-vehicle device status includes the degree of window opening / closing, air conditioning fan speed, etc.; the driving status includes low-speed driving and high-speed driving, etc.; the current status also includes the status of the in-vehicle software, such as whether multimedia is in use, etc.; available resources include in-vehicle address book, in-vehicle map information, and other usable resources; historical dialogue information includes the user's historical request text and historical dialogue scene values.
[0039] Step S1202: Based on the current dialogue information, current state, available resources and historical dialogue information, determine the target intelligent agent label from a set of preset intelligent agent labels.
[0040] In this embodiment, the user's historical dialogue information can serve as the basis for determining whether to provide the same in-vehicle system function as the current dialogue information. The vehicle's current state provides a basis for cross-scenario considerations, and available resources provide a basis for cross-functional considerations. Therefore, determining the target agent label also requires considering the current state, available resources, and historical dialogue information to make the determined target agent label more closely match the user's intent and the vehicle's current state.
[0041] Optionally, step S1202 includes: The first step is to match the target keywords corresponding to the current dialogue information from the preset keyword template library, and determine the first agent label and the first matching degree corresponding to the target keywords. The first matching degree is the matching degree between the current dialogue information and the first agent label.
[0042] In this embodiment, a matching module can be pre-built using Redis (a key-value database) and Nacos (a dynamic configuration platform). Redis, as a high-performance cache, stores commonly used keyword matching templates and dictionaries, with each keyword matching template and agent label stored in a one-to-one correspondence. Nacos is used to dynamically obtain the latest template configuration information by configuring hot-update content for different environments and services. The current dialogue information is input into the matching template, which then performs matching and outputs the target keyword, along with the corresponding agent label, denoted as the first agent label. Based on the matching length and continuity between the current request text and the keyword matching template, the matching degree between the current dialogue information and the first agent label is calculated and denoted as the first matching degree.
[0043] Optional, the first step includes: A template matching algorithm is used to match the first matching keyword corresponding to the current dialogue information from the first keyword template library, and the first similarity between the current dialogue information and the first matching keyword is determined. A fuzzy matching algorithm is used to match the second matching keyword corresponding to the current dialogue information from the second keyword template library, and the second similarity between the current dialogue information and the second matching keyword is determined. If the first similarity is greater than the second similarity, the first matching keyword is used as the target keyword. If the first similarity is greater than or equal to the second similarity, the second matching keyword is used as the target keyword.
[0044] This can be understood as the matching module comprising a template matching module and a fuzzy matching module. The template matching module extracts the request keywords from the current request text in the current dialogue information, then calculates the similarity between each first keyword template in the first keyword template library related to the current dialogue scenario value and the request keywords, and takes the first keyword template with the highest similarity as the first matching keyword, and the corresponding similarity as the first similarity.
[0045] The fuzzy matching template calculates the etymological similarity, edit distance, and continuous character matching degree between the current request text and each second keyword template in the second keyword template library that is related to the current dialogue scenario value. Then, it performs a weighted sum to obtain the similarity between the current request text and each second keyword template. The second keyword template with the highest similarity is taken as the second matching keyword, and the corresponding similarity is taken as the second similarity.
[0046] Finally, the first similarity and the second similarity are compared, and the larger of the two is taken as the target similarity. The matching keyword corresponding to the target similarity is taken as the target keyword. That is, if the first similarity is greater than the second similarity, the first matching keyword is taken as the target keyword and the first similarity is taken as the target similarity; if the first similarity is greater than or equal to the second similarity, the second matching keyword is taken as the target keyword and the second similarity is taken as the target similarity.
[0047] In this embodiment, the combined use of the template matching module and the fuzzy matching module can support more generalized user intent scenarios, enabling preliminary analysis of user intent and reducing the pressure of subsequent analysis.
[0048] Optionally, the method also includes: The larger of the first similarity and the second similarity is taken as the target similarity; if the target similarity is greater than the preset similarity threshold, the matching function corresponding to the target keyword is taken as the function material.
[0049] In this embodiment, the keyword template library stores keyword templates in a one-to-one correspondence with functional functions, and different keyword templates may correspond to different functional functions. Therefore, when the target similarity corresponding to the target keyword is greater than the preset similarity threshold, it indicates that the user's intent is very close to the meaning of the target keyword. At this time, the matching module not only outputs the first agent label, but also outputs the corresponding functional function, denoted as the matching functional function, and uses this matching functional function as functional function material. At this time, the functional function material includes not only the initially selected functional function, but also the matching functional function, which can provide more references for subsequently determining the target functional function.
[0050] The second step is to determine the second agent label and the second matching degree from the preset labels of multiple agents based on the current dialogue information, current state, available resources and historical dialogue information. The second matching degree is the matching degree between the current dialogue information, current state, available resources and historical dialogue information and the second agent label.
[0051] In this embodiment of the application, a pre-classification module can be pre-built. The current dialogue information, current state, available resources and historical dialogue information are input into the pre-classification module. The pre-classification module outputs a second agent label and the matching degree between the current dialogue information, current state, available resources and historical dialogue information and the second agent label, i.e., the second matching degree.
[0052] The pre-classification module can be a lightweight model with fewer than 1B of parameters. This lightweight model improves the inference speed of the pre-classification module, has low power consumption, and allows for local deployment and customization. A large amount of historical dialogue information, historical states, available resources, and corresponding agent labels can be used as a sample dataset. The model is trained and optimized using this sample dataset to obtain a lightweight model with satisfactory accuracy and precision. The specific training and optimization process can utilize existing technologies and will not be elaborated upon here.
[0053] The method described in this application can improve task execution efficiency and accuracy: the refined task planning of the pre-classification module and the efficient recall mechanism based on offline MongoDB + online Elasticsearch ensure the orderly execution of the task and the accuracy of data acquisition. The system can quickly determine the key steps and required resources of the task, and accurately retrieve relevant functional materials from the functional function library, thereby generating high-quality responses and accurately calling the corresponding functional services. This avoids invalid searches or erroneous executions that may occur in traditional systems, greatly improving the efficiency and accuracy of task execution and saving users' time and effort.
[0054] Third step: If the first matching degree is greater than the second matching degree, then the first agent label is used as the target agent label.
[0055] In this embodiment of the application, if the first matching degree is greater than the second matching degree, it means that the first agent tag matched by the matching module is closer to the user's intention, and the first agent tag is used as the target agent tag.
[0056] Fourth step: If the first matching degree is less than or equal to the second matching degree, then the second agent label is used as the target agent label.
[0057] In this embodiment, if the first matching degree is less than or equal to the second matching degree, it means that the second agent label separated by the pre-classification module is closer to the user's intention, and the second agent label is used as the target agent label.
[0058] Optionally, step S130 includes: Step S1301: Determine the correlation between the current dialogue information and the target agent label and each function in the preset function library; In this embodiment, the relevance can be determined by weighted summation based on the current dialogue information and the target agent's label, as well as the label matching degree and fuzzy text similarity of each function in the preset function library.
[0059] Optionally, step S1301 includes: The current dialogue information and the target agent label are converted into retrieval vector data; the vector similarity between the retrieval vector data and the function vector data corresponding to each function in the preset function library is calculated; based on the vector similarity, the relevance between the current dialogue information and the target agent label and each function is determined.
[0060] In this embodiment, the function functions in the function library have corresponding function vector data, which are stored one-to-one in a MongoDB database. Different retrieval vectors are constructed in the recall module using a pre-built embedding model, and a multi-category retrieval index is built for function or data fragment retrieval. Relevant data from the cockpit domain can be used as a sample dataset to train and optimize the embedding model; the specific construction steps are existing technology and will not be elaborated here.
[0061] Specifically, the current dialogue information and target agent labels can be converted into retrieval vector data first. The vector similarity between the retrieval vector data and the function vector data corresponding to each function in the preset function library can be calculated. The closer the retrieval vector data and the function vector data are in space, that is, the greater the vector similarity, the more semantically similar the current dialogue information and target agent labels are to the definitions of the corresponding function. Then, based on the label matching degree, fuzzy text similarity and vector similarity between the current dialogue information and target agent labels and each function in the preset function library, a weighted sum is calculated to determine the relevance between the current dialogue information and target agent labels and each function.
[0062] Step S1302: Select function functions with a relevance greater than the preset relevance threshold as initial function functions.
[0063] In this embodiment of the application, after the recall module calculates the relevance, it will output the function with a relevance greater than the relevance threshold as the initial function.
[0064] In this embodiment, if the user inputs voice, the current dialogue information also includes the voice zone location, i.e., the location where the voice was emitted, such as the driver's seat or the rear seat. Finally, the generated function parameters also include the control location; for example, if the voice zone location is the driver's seat, then the air conditioning corresponding to the driver's seat will be turned on.
[0065] In this embodiment, the target function output by the large language model is a refined function. This can be understood as follows: the large language model first obtains a candidate function based on current dialogue information, historical dialogue information, and function material; then, based on the definition of the candidate function and the answer resource generated from the current request text, it refines the candidate function and the response text to obtain the target function, ensuring the accuracy of the function, the completeness of the parameters, and the integrity and logic of the information.
[0066] For example, if the system recognizes that the user's voice input is from a child, and the resulting candidate function conflicts with the child's identity, the candidate function needs to be refined to obtain a target function that matches the child's identity. For instance, if a child wants to open a window, but for safety reasons, the window-opening function cannot be executed, a text response indicating that the window cannot be opened can be generated. This text response can then be converted into speech content for output.
[0067] The method in this application embodiment can improve user experience and interaction convenience: through accurate intent matching and natural and fluent response generation, users can interact with the intelligent cockpit system in a more natural and convenient way in the in-vehicle environment. Whether querying vehicle information, controlling vehicle functions, or obtaining entertainment information, accurate responses can be obtained quickly, reducing the complexity and distraction of user operations during driving and improving driving safety. Enhanced system intelligence and adaptability: Based on the self-developed intelligent cockpit large language model and the collaborative work of various modules, the system can deeply understand and adapt to various complex needs and semantic changes in in-vehicle scenarios. Compared with traditional in-vehicle intelligent systems, the intelligent agent of this invention can handle more diverse and personalized user commands. For fuzzy or incomplete commands, it can also generate reasonable responses through fuzzy matching and planning, demonstrating a higher level of intelligence and adaptability, better meeting the needs of different users in different driving scenarios. Facilitates system expansion and optimization: The use of multiple databases such as Redis, Nacos, MongoDB, and Elasticsearch increases the system's scalability and independence. When it is necessary to update the knowledge base, optimize matching rules, or upgrade the model, the corresponding modules can be easily operated on without causing a major impact on the entire system. Reduced system resource consumption and costs: By utilizing small models for task planning, along with Redis's caching mechanism and efficient recall algorithm, the system's consumption of computing and storage resources is effectively reduced. This reduces reliance on hardware and associated costs while maintaining system performance.
[0068] Based on the same concept, embodiments of the present invention also provide a processing device for intelligent cockpit dialogue. Figure 2 This is a structural block diagram of a smart cockpit dialogue processing device provided in an embodiment of this application, as shown below. Figure 2 As shown, the device 200 includes an acquisition module 201, a first determination module 202, a retrieval module 203, a material module 204, a second determination module 205, and a control module 206.
[0069] The acquisition module 201 is used to acquire the current dialogue information, which includes the current request text and the current dialogue scenario value. The first determining module 202 is used to determine the target intelligent agent label from a plurality of preset intelligent agent labels based on the current dialogue information. The retrieval module 203 is used to retrieve one or more initially selected function functions from a preset function function library based on the current dialogue information and the target intelligent agent's label; Material module 204 is used to use the initially selected function as function material; The second determining module 205 is used to input the current dialogue information and function material into the pre-built large language model to obtain the response text or function parameters of the target function output by the large language model. The control module 206 is used to call the vehicle system function corresponding to the target function function if the output of the large language model is a function parameter, and adjust the parameters of the vehicle system function according to the function parameter.
[0070] Optionally, the first determining module 202 includes: The acquisition unit is used to acquire the current status of the vehicle, the available resources of the in-vehicle system, and historical dialogue information. The current status includes the status of the in-vehicle equipment and the driving status. The first determining unit is used to determine the target agent label from a set of preset agent labels based on the current dialogue information, current state, available resources and historical dialogue information.
[0071] Optionally, the determining unit includes: The first determining subunit is used to match the target keyword corresponding to the current dialogue information from the preset keyword template library, and determine the first agent label and the first matching degree corresponding to the target keyword. The first matching degree is the matching degree between the current dialogue information and the first agent label. The second determining subunit is used to determine the second agent label and the second matching degree from the preset labels of multiple agents based on the current dialogue information, the current state, the available resources and the historical dialogue information. The second matching degree is the matching degree between the current dialogue information, the current state, the available resources and the historical dialogue information and the second agent label. The third determining subunit is used to take the first agent label as the target agent label if the first matching degree is greater than the second matching degree. The fourth determining subunit is used to take the second agent label as the target agent label if the first matching degree is less than or equal to the second matching degree.
[0072] Optionally, the first determined subunit is also used for: A template matching algorithm is used to match the first matching keyword corresponding to the current dialogue information from the first keyword template library, and to determine the first similarity between the current dialogue information and the first matching keyword. A fuzzy matching algorithm is used to match the second matching keyword corresponding to the current dialogue information from the second keyword template library, and to determine the second similarity between the current dialogue information and the second matching keyword. If the first similarity score is greater than the second similarity score, then the first matching keyword will be used as the target keyword. If the first similarity is greater than or equal to the second similarity, then the second matching keyword will be used as the target keyword.
[0073] Optionally, the first determined subunit is also used for: The larger of the first similarity and the second similarity is taken as the target similarity. If the target similarity is greater than the preset similarity threshold, the matching function corresponding to the target keyword will be used as the function material.
[0074] Optionally, the retrieval module 203 includes: The second determining unit is used to determine the correlation between the current dialogue information and the target agent label and each function in the preset function function library; The third determining unit is used to select function functions with a relevance greater than a preset relevance threshold as initial function functions.
[0075] Optionally, the second determining unit is also used for: Convert the current dialogue information and the target agent's label into retrieval vector data; Calculate the vector similarity between the retrieved vector data and the function vector data corresponding to each function in the preset function library; Based on vector similarity, the correlation between the current dialogue information and the target agent label and each functional function is determined.
[0076] It is understood that the device provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.
[0077] This invention also provides an electronic device that may include a processor and a memory, wherein the processor and the memory may be interconnected via a bus or other means.
[0078] The processor can be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application, or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or other chips, or combinations of the above types of chips.
[0079] Memory may include mass storage for data or instructions. For example, and not limitingly, memory may include hard disk drives (HDDs), floppy disk drives, flash memory, optical disks, magneto-optical disks, magnetic tape, or Universal Serial Bus (USB) drives, or combinations of two or more of these. Where appropriate, memory may include removable or non-removable (or fixed) media. Where appropriate, memory may be internal or external to an electronic device. In a particular embodiment, memory may be non-volatile solid-state memory.
[0080] In one instance, the memory may be read-only memory (ROM). In one instance, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.
[0081] The processor reads and executes computer program instructions stored in the memory to implement any of the intelligent cockpit dialogue processing methods in the above embodiments.
[0082] In one example, the electronic device may further include a communication interface and a bus. The processor, memory, and communication interface are connected via the bus to communicate with each other. The communication interface is primarily used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of this application. Where appropriate, the bus may include one or more buses.
[0083] Furthermore, in conjunction with the intelligent cockpit dialogue processing methods in the above embodiments, this invention can be implemented using a computer-readable storage medium. This computer-readable storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the intelligent cockpit dialogue processing methods in the above embodiments.
[0084] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The storage medium can be read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium can also include combinations of the above types of memory.
[0085] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: This application provides a method, apparatus, device, and medium for processing intelligent cockpit dialogue. The method involves acquiring current dialogue information, obtaining user input commands and dialogue scenarios, including the current request text and the current dialogue scenario value. Based on the current dialogue information, a target intelligent agent label is determined from a set of preset intelligent agent labels to initially analyze the user's intent. Based on the current dialogue information and the target intelligent agent label, one or more preliminary function selections are retrieved from a preset function library to further analyze ways to resolve the user's intent. Functional function materials are constructed, including the preliminary function selections, to obtain several possible user intents. The current dialogue information and functional function materials are input into a pre-constructed large language model to obtain the response text output by the large language model or the function parameters of the target function. The most likely user intent is analyzed, and a response is made. If the large language model outputs function parameters, the corresponding in-vehicle system function is called, and the parameters of the in-vehicle system function are adjusted according to the function parameters. If the user wants to control the in-vehicle system function, the corresponding in-vehicle system function is executed. This method can first perform preliminary processing and intent analysis on the current dialogue information between the user and the vehicle system to obtain functional materials related to the user's intent. Then, the current dialogue information and functional materials are input into a large language model to obtain the response text or functional parameters of the target functional function output by the large language model. It can also call the vehicle system function corresponding to the target functional function to realize intelligent interaction between the user and the vehicle system. Only one call to the large language model is needed, which improves the end-to-end response speed and can respond to user requests more quickly.
[0086] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.
[0087] Similarly, it should be understood that, in order to simplify this disclosure and aid in understanding one or more of the various aspects of the invention, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof. However, this method of disclosure should not be construed as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into this detailed description, wherein each claim itself is a separate embodiment of the invention.
[0088] It should be noted that the above embodiments are illustrative of the invention and not restrictive, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names.
Claims
1. A method for processing dialogue in an intelligent cockpit, characterized in that, The method includes: Obtain current dialogue information, which includes the current request text and the current dialogue scenario value; Based on the current dialogue information, the target intelligent agent label is determined from a set of preset intelligent agent labels; Based on the current dialogue information and the target agent's label, retrieve one or more preliminary function selections from a preset function library; The initially selected function will be used as the function source material. The current dialogue information and the functional material are input into a pre-built large language model to obtain the response text or functional parameters of the target functional function output by the large language model. If the output of the large language model is the function parameter, then the vehicle system function corresponding to the target function is called, and the parameters of the vehicle system function are adjusted according to the function parameter.
2. The intelligent cockpit dialogue processing method according to claim 1, characterized in that, The step of determining the target agent tag from a set of preset agent tags based on the current dialogue information includes: The vehicle's current status, the available resources of the in-vehicle system, and historical dialogue information are obtained. The current status includes the status of the in-vehicle equipment and the driving status. Based on the current dialogue information, the current state, the available resources, and the historical dialogue information, the target agent label is determined from a set of preset agent labels.
3. The intelligent cockpit dialogue processing method according to claim 2, characterized in that, The step of determining the target agent tag from a set of preset agent tags based on the current dialogue information, the current state, the available resources, and the historical dialogue information includes: The target keyword corresponding to the current dialogue information is matched from the preset keyword template library, and the first agent label and the first matching degree corresponding to the target keyword are determined. The first matching degree is the matching degree between the current dialogue information and the first agent label. Based on the current dialogue information, the current state, the available resources, and the historical dialogue information, a second agent label and a second matching degree are determined from a set of preset agent labels. The second matching degree is the matching degree between the current dialogue information, the current state, the available resources, and the historical dialogue information and the second agent label. If the first matching degree is greater than the second matching degree, then the first agent label is used as the target agent label; If the first matching degree is less than or equal to the second matching degree, then the second agent label is used as the target agent label.
4. The intelligent cockpit dialogue processing method according to claim 3, characterized in that, The keyword template library includes a first keyword template library and a second keyword template library. Matching the target keyword corresponding to the current dialogue information from the preset keyword template library includes: A template matching algorithm is used to match the first matching keyword corresponding to the current dialogue information from the first keyword template library, and to determine the first similarity between the current dialogue information and the first matching keyword. A fuzzy matching algorithm is used to match the second matching keyword corresponding to the current dialogue information from the second keyword template library, and to determine the second similarity between the current dialogue information and the second matching keyword. If the first similarity is greater than the second similarity, then the first matching keyword is taken as the target keyword; If the first similarity is greater than or equal to the second similarity, then the second matching keyword is used as the target keyword.
5. The intelligent cockpit dialogue processing method according to claim 4, characterized in that, The method further includes: The larger of the first similarity and the second similarity is taken as the target similarity; If the target similarity is greater than the preset similarity threshold, then the matching function corresponding to the target keyword will be used as the function material.
6. The intelligent cockpit dialogue processing method according to claim 1, characterized in that, The step of retrieving one or more initially selected function functions from a preset function function library based on the current dialogue information and the target agent label includes: Determine the correlation between the current dialogue information and the target agent label and each function in the preset function library; Functions with a relevance greater than a preset relevance threshold are selected as the initial function functions.
7. The intelligent cockpit dialogue processing method according to claim 6, characterized in that, Determining the relevance of the current dialogue information and the target agent label to each function in the preset function library includes: The current dialogue information and the target agent label are converted into retrieval vector data; Calculate the vector similarity between the retrieved vector data and the function vector data corresponding to each function in the preset function library; Based on vector similarity, the correlation between the current dialogue information and the target agent label and each functional function is determined.
8. A processing device for intelligent cockpit dialogue, characterized in that, The device includes: The acquisition module is used to acquire current dialogue information, which includes the current request text and the current dialogue scenario value; The first determining module is used to determine the target intelligent agent label from a plurality of preset intelligent agent labels based on the current dialogue information; The retrieval module is used to retrieve one or more initially selected function functions from a preset function function library based on the current dialogue information and the target intelligent agent label; The material module is used to use the initially selected function as function material; The second determining module is used to input the current dialogue information and the functional material into a pre-constructed large language model to obtain the response text or functional parameters of the target functional function output by the large language model. The control module is used to call the vehicle system function corresponding to the target function function if the output of the large language model is the function parameter, and adjust the parameters of the vehicle system function according to the function parameter.
9. An electronic device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method of any one of claims 1-7.