Vehicle enterprise APP intelligent assistant navigation method and system based on large model

Abstract

The invention provides a vehicle enterprise APP intelligent assistant navigation method and system based on a large model. The method relates to collection of user navigation requirements and identification of key entities and user intentions by using a large language model. And when the navigation intention is identified, generating a navigation route through map data and a path planning algorithm in combination with the current position and the target position. And when the vehicle control intention is identified, remotely controlling the vehicle through a vehicle control interface. And when the intention is identified as the information query intention, relevant information is retrieved and displayed from the vehicle information base. And collecting user feedback, an optimization model and a navigation strategy. The navigation demand information comprises text and voice information. The navigation requirement understanding of the user is improved through the large language model, tedious operation is avoided, and the access efficiency and the user experience are improved. Personalized and real-time recommendation is realized, real-time information is provided according to requirements and environments, and travel route planning is optimized. Dialogue management supports multiple rounds of dialogues and deep interaction, complex requirements are effectively processed, and the ability to deal with complex scenes is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent vehicle navigation, and particularly to a navigation method and system for an intelligent assistant of a vehicle enterprise APP based on a large model. Background Art

[0002] In the current market environment, the in-vehicle application programs (hereinafter referred to as vehicle enterprise APPs) launched by many automobile manufacturers mainly provide many practical functions such as real-time query of vehicle status, reservation of maintenance services, query of repair records, and location search of charging piles. Although these functions meet the basic needs of vehicle owners to a certain extent, in terms of the user navigation function, vehicle enterprise APPs generally face some challenges and deficiencies. First of all, the traditional navigation method requires vehicle owners to make cumbersome selections through layers of menus or search by categories. Such an operation process is relatively complex and lacks intelligent automatic recommendation and guidance, resulting in an unsatisfactory user experience. Secondly, the existing vehicle enterprise APPs often lack personalized and real-time recommendations in the navigation function and cannot provide active guidance services according to the specific needs or context of vehicle owners, which greatly limits the improvement of the user experience. Finally, the vehicle enterprise APPs are also insufficient in terms of dialogue management ability. They usually cannot support multi-round conversations and in-depth interactions, which makes them unable to handle complex requirements raised by vehicle owners effectively and unable to provide effective solutions. Summary of the Invention

[0003] The present invention aims to at least solve the technical problem of low access efficiency in the existing technology, and particularly innovatively proposes a navigation method and system for an intelligent assistant of a vehicle enterprise APP based on a large model.

[0004] To achieve the above object of the present invention, the present invention provides a navigation method for an intelligent assistant of a vehicle enterprise APP based on a large model, and the method includes: S1. Collect the navigation requirement information of the user, obtain key entities by using a large language model based on the navigation requirement information, and identify the user intention based on the key entity information; S2. When the identified user intention is a navigation intention, generate a navigation route according to the current location and the target location by using map data and a path planning algorithm; S3. When the identified user intention is a vehicle control intention, communicate with the vehicle through a vehicle control interface based on the vehicle control intention of the user, and remotely control the vehicle; S4. When the identified user intention is an information query intention, retrieve relevant information from a preset vehicle information library according to the query content and display it to the user; S5. Collect the feedback of the user, and optimize the large language model and the navigation strategy based on the feedback.

[0005] As an alternative embodiment of the present invention, optionally, the navigation requirement information in step S1 includes text information and voice information; When the information input by the user is voice information, a wake-up word is set, and the voice information is converted into text information by using a voice recognition method, and then the key entities are extracted and the user intention is recognized by using a large language model.

[0006] As an alternative embodiment of the present invention, optionally, generating a navigation route by using map data and a path planning algorithm in step S2 includes: S201. Based on the current location and the target location, obtain the corresponding road network information and traffic condition information from the map data; S202. Based on the road network information and the traffic condition information, calculate the optimal path from the current location to the target location by using a path planning algorithm; S203. Use the optimal path as the navigation route to display to the user, and update the traffic condition information in real time.

[0007] As an alternative embodiment of the present invention, optionally, remotely controlling the vehicle in step S3 includes: S301. Based on the vehicle control intention of the user, establish a communication connection with the vehicle through a vehicle control interface; S302. Generate a remote control instruction according to the vehicle control intention of the user, and remotely control the vehicle based on the remote control instruction; S303. Real-time feedback the execution result of the remote control to the user.

[0008] As an alternative embodiment of the present invention, optionally, the method further includes: S6. Record the historical navigation requirement information, and interact with the user based on the historical navigation requirement information and the current navigation requirement information.

[0009] As an alternative embodiment of the present invention, optionally, when the recognized user intention in step S4 is an information query intention, retrieving relevant information from a preset vehicle information library according to the query content and displaying it to the user includes: S401. Based on the query content of the user, retrieve vehicle information entries related to the user's query content from a preset vehicle information library; S402. Screen and sort the retrieved vehicle information entries, and determine the display order to the user according to the relevance and importance of the information to the query content.

[0010] On the other hand, the present invention also provides a navigation system for a smart assistant of a car company APP based on a large model, the system is based on the navigation method of the smart assistant of the car company APP based on a large model; the system further includes: A user interaction module, configured to receive navigation requirement information of a user, display navigation routes, remote control results, and information query results; A speech recognition module, connected to the user interaction module, configured to convert speech information input by the user into text information; A natural language processing module, connected to the speech recognition module, configured to identify user intentions by using the text information; A map data module, connected to the natural language processing module, configured to store and provide map data; A vehicle control interface module, connected to the natural language processing module, configured to communicate with a vehicle; A vehicle information database module, connected to the natural language processing module, configured to store vehicle-related information.

[0011] Advantages of the present invention: By introducing a large language model, the present invention realizes efficient understanding and intelligent recommendation of users' navigation requirements. Compared with the traditional navigation method of car manufacturers' APPs, first, the present invention deeply analyzes the text or speech information input by the user through the large language model, accurately identifies the user's intentions, thereby avoiding the user from searching for the required functions through cumbersome menu selections or searches, greatly improving the access efficiency. This intelligent automatic recommendation and guidance significantly enhance the user experience. Second, the present invention realizes personalized and real-time recommendations in the navigation function. According to the specific needs or context of the user, the present invention can actively provide real-time information such as navigation routes and traffic conditions to help the user better plan the itinerary. This personalized service not only enhances the user experience but also increases the user's stickiness to the car manufacturers' APP. In addition, the present invention also performs well in dialogue management. It supports multi-round conversations and in-depth interactions, can handle complex requirements proposed by the user, and provide effective solutions. This powerful dialogue management ability enables the present invention to handle various complex scenarios with ease.

[0012] Additional aspects and advantages of the present invention will be partly given in the following description, partly will become obvious from the following description, or be understood through the practice of the present invention. Description of the Drawings

[0013] The above and / or additional aspects and advantages of the present invention will become obvious and easy to understand from the description of the embodiments in conjunction with the following drawings, where: Figure 1 is a flowchart of a navigation method for an intelligent assistant of a car manufacturers' APP based on a large model in Embodiment 1 of the present invention; Figure 2 is a schematic flowchart of the interaction between an intelligent assistant navigation system of a car manufacturers' APP based on a large model and a user in Embodiment 2 of the present invention; Figure 3 It is a schematic diagram of the voice recognition process of the voice recognition module of a large model-based vehicle enterprise APP intelligent assistant navigation system according to Embodiment 2 of the present invention; Figure 4 It is a schematic diagram of the processing process of the natural language processing module of a large model-based vehicle enterprise APP intelligent assistant navigation system according to Embodiment 2 of the present invention; Figure 5 It is a schematic diagram of the background interaction process of a large model-based vehicle enterprise APP intelligent assistant navigation system according to Embodiment 2 of the present invention; Figure 6 It is a schematic diagram of the structure of a large model-based vehicle enterprise APP intelligent assistant navigation system according to Embodiment 2 of the present invention. Detailed implementation manners

[0014] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation to the present invention.

[0015] Embodiment 1

[0016] As Figure 1 shown, a large model-based vehicle enterprise APP intelligent assistant navigation method, the method includes: S1. Collect the navigation requirement information of the user, obtain the key entities based on the navigation requirement information by using a large language model, and identify the user intention based on the key entity information; It should be noted that in step S1, the navigation requirement information includes voice information and text information. The voice information is input through voice interaction between the user and the intelligent assistant, and the text information is input through the input box on the vehicle enterprise APP. The large language model deeply analyzes the voice or text information input by the user, extracts the key entities therein, such as place names, time requirements, etc., and further identifies the user's navigation intention based on these key entities, such as finding the optimal route, querying traffic conditions, etc. In this way, the present invention can accurately understand the user's navigation requirements.

[0017] S2. When the identified user intention is a navigation intention, generate a navigation route according to the current location and the target location by using map data and a path planning algorithm; It should be noted that when the recognized user intention in step S2 is a navigation intention, the system will first obtain the user's current location and destination location. These two location information can be obtained through the user's input (such as voice commands or direct selection on the APP), or automatically obtained through positioning technology. Then, the system will utilize the map data stored in the server, which includes detailed information of roads, traffic rules, road restrictions, etc., as well as real-time traffic condition information, such as traffic congestion, accidents, construction, etc. Based on this information, the system will adopt an advanced path planning algorithm to calculate the optimal path from the current location to the destination location. Finally, the system will display the calculated optimal path as the navigation route to the user and update the traffic condition information in real time during the navigation process so that the user can adjust the itinerary in a timely manner.

[0018] S3. When the recognized user intention is a vehicle control intention, based on the user's vehicle control intention, communicate with the vehicle through the vehicle control interface to remotely control the vehicle; It should be noted that when the recognized user intention in step S3 is a vehicle control intention, the system will communicate with the vehicle through the vehicle control interface according to the user's specific vehicle control requirements, such as turning on the air conditioner, adjusting the seat, starting the vehicle, etc. This vehicle control interface is established based on the API (Application Programming Interface) provided by the vehicle manufacturer, which allows the APP to perform data transmission and instruction interaction with the vehicle. The system will generate corresponding remote control instructions according to the user's vehicle control intention and send them to the vehicle through this interface. After receiving the instructions, the vehicle will perform the corresponding operations and feedback the execution results to the APP through the interface. The APP will then display this execution result to the user in real time so that the user can understand the vehicle status and control results. In this way, the present invention realizes the remote control of the vehicle and provides a more convenient and intelligent vehicle use experience for users.

[0019] S4. When the recognized user intention is an information query intention, retrieve relevant information from the preset vehicle information database according to the query content and display it to the user; It should be noted that when the recognized user intention in step S4 is an information query intention, the system will retrieve relevant information from the preset vehicle information database according to the user's query content, such as vehicle status, maintenance reminder, fault information, etc. This vehicle information database is established based on the vehicle data provided by the vehicle manufacturer and the user's vehicle use history data, which includes detailed information of the vehicle, maintenance records, fault history, etc. The system will screen and sort the retrieved information and determine the display order for the user according to the relevance and importance of the information to the query content. In this way, the present invention can provide accurate, timely and useful vehicle information for users, help users better understand the vehicle status and make reasonable vehicle use decisions.

[0020] S5. Collect the feedback from users and optimize the large language model and navigation strategy based on the feedback.

[0021] It should be noted that in step S5, the system continuously collects the feedback from users during the usage process. This feedback includes the user's satisfaction with the navigation route, the evaluation of the response speed of vehicle remote control, the view on the accuracy of information query results, etc. The system will sort out and analyze this feedback information to find out the existing problems and deficiencies. Then, based on this feedback information, the system will train and optimize the large language model to improve its parsing ability for user input and the accuracy of intention recognition. At the same time, the system will also adjust and optimize the navigation strategy to provide a navigation service that better meets the user's needs. In this way, the present invention can continuously learn and progress to provide a more intelligent, convenient and efficient navigation service for users.

[0022] In summary, this embodiment provides a navigation method for an intelligent assistant of a car company APP based on a large model. This method collects the navigation requirement information of users, uses a large language model for in-depth parsing and intention recognition, generates a navigation route, performs vehicle remote control or queries vehicle information according to the recognized user intention, and collects user feedback to optimize the large language model and navigation strategy, thereby realizing an intelligent, convenient and efficient navigation service. This method not only improves the efficiency, accuracy and user experience of navigation, but also enhances the intelligent level and service quality of the system through continuous learning and optimization.

[0023] As an optional embodiment of the present invention, optionally, in step S1, the navigation requirement information includes text information and voice information; When the user inputs voice information, a wake-up word is set, and the voice information is converted into text information by using a voice recognition method, and then the large language model is used to extract key entities and recognize the user's intention.

[0024] It should be noted that when the user inputs voice information, the system first activates the voice recognition function through a preset wake-up word. Once the wake-up word is recognized, the system will start the voice recognition module to accurately convert the user's voice information into text information. This conversion process relies on voice recognition technology, which can overcome challenges such as noise and accent differences to ensure that the voice information is accurately converted into text. Subsequently, the system will input the converted text information into the large language model for key entity extraction and user intention recognition. With its powerful language understanding and analysis capabilities, the large language model can quickly and accurately identify key entities in the text input by the user, such as locations, times, etc., and further analyze the user's navigation intention. In this way, the present invention not only supports text input, but also can efficiently process voice input, providing a more flexible and diverse interaction method for users.

[0025] As an alternative embodiment of the present invention, optionally, the expression for converting the voice information into text information is: , , , , , , ; wherein, y(t) represents the noisy speech signal (time-domain continuous signal) collected by the microphone; t represents the time variable (unit: second); x(t) represents the original speech signal (without noise interference); n(t) represents the environmental noise signal (usually modeled as Gaussian white noise); y filt (t) represents the filtered enhanced speech signal; * represents the convolution operator; h(t) represents the impulse response function of the filter (such as Wiener filter or Kalman filter); y(τ) represents the amplitude of the original noisy speech signal at time τ ; τ represents the integral dummy variable (time offset); h(t-τ) represents the gain of the impulse response function at the time difference t-τ ; Y m,k represents the complex spectral coefficient of the m th frame and the k th frequency point; N represents the frame length (the number of samples per frame, typical value: 25 ms corresponds to 200 - 400 samples); n represents the number of frames; m represents the frame index ( m =0, 1,..., M−1, a total of M frames); HIndicates the frame shift (the number of samples between adjacent frames, usually taking 1 / 2 or 1 / 3 of the frame length N); w(n) Indicates the window function; e -j2πkn / N Indicates the complex rotation factor, specifically representing the rotation of a complex number in the complex plane, with an angle of -j2π kn / N , and the negative sign indicates counterclockwise rotation; MFCC m Indicates the m Mel-frequency cepstral coefficient vector of the DCT ( ) indicates the discrete cosine transform (converting the spectrum to cepstral coefficients); b i Indicates the i th Mel-filter bank boundary (unit: Hz); M i Indicates the i th Mel-filter weight (triangle filter response); h m Indicates the encoded hidden state vector; Encoder ( ) indicates the encoder (such as Transformer or LSTM); P (w) indicates the joint probability of the output text sequence w; L Indicates the length of the output text sequence (number of tokens); w l Indicates the l th token (such as Chinese character, pinyin, or sub-word unit); Indicates the finally optimized text output; P ASR (w) indicates the word sequence probability output by the speech recognition model; λ Indicates the language model weight coefficient (hyperparameter, balancing the contributions of ASR and LM); P LM (w) indicates the grammar correction probability given by the language model (such as n-gram or RNN-LM).

[0026] As an alternative embodiment of the present invention, optionally, the expression for obtaining the key entity using the large language model in step S1 is: ; Among them, Entities represents the final set of extracted entities, which is used to store all entities (such as person names, locations, etc.) identified from the text; (w l * ,tag l * ) represents an entity pair, which is composed of the token w l * and its label tag l * Specifically, it means that the l th token in the text is identified as the entity type tag l * (such as "Beijing" being marked as a location); represents finding the label sequence that maximizes the product of probabilities among all label sequences tag to determine the optimal label sequence by maximizing the conditional probability; P ( tag l | w * , tag 1:l-1 ) represents the probability that the * th entity belongs to the label tag 1:l-1 under the condition of the given text w l and the historical label tag l ; tag 1:l-1 represents the label sequence from the l th entity to the l - 1th entity.

[0027] As an alternative embodiment of the present invention, optionally, the expression for identifying the user intention based on the key entity information in step S1 is: ; Among them, I * represents the optimal user intention, specifically representing the most likely intention category predicted by the model (such as "play music"); represents finding the category with the maximum probability among all intention categories C inc k , which is used to determine the optimal intention by maximizing the probability; C represents the set of all intention categories and is used to define the output space of intention recognition; represents an intention category c k the corresponding weight vector, specifically mapping the feature vector to the score of the intention category; L represents the length of the input entity sequence; represents the aggregated entity feature vector. Specifically, through average pooling, multiple entity features are fused into a global feature; e l represents the l - th word embedding vector of the entity, which is the semantic information of the entity; t l represents the l - th label embedding vector of the entity (such as the vector of "location"), which is the type information of the entity; represents the vector concatenation operation, concatenating the word embedding and the label embedding into an entity feature; represents an intention category c k the corresponding bias term; K represents the number of categories in the classification task.

[0028] As an alternative embodiment of the present invention, optionally, generating a navigation route using map data and a path planning algorithm in step S2 includes: S201. Based on the current location and the target location, obtain the corresponding road network information and traffic condition information from the map data; It should be noted that in step S201, the system will first use high - precision map data to determine all possible paths between the current location and the target location. These paths will be preliminarily screened based on the road network information to ensure that they are all actual feasible routes. Then, the system will consider real - time traffic condition information, such as road conditions, congestion situations, traffic accidents, etc., to further optimize these paths. By comprehensively evaluating factors such as travel time, distance, and comfort of different paths, the system can recommend the best navigation route for the user. In addition, the system will continuously monitor changes in traffic conditions and automatically adjust the navigation route when necessary to avoid potential congestion or delays, ensuring that the user can reach the destination smoothly and efficiently.

[0029] S202. Based on the road network information and traffic condition information, use a path planning algorithm to calculate the optimal path from the current location to the target location; It should be noted that in step S202, the system will use a path planning algorithm, such as the A* algorithm, Dijkstra algorithm or its improved version, to calculate the optimal path from the current location to the target location. These algorithms will comprehensively consider multiple factors such as the length of the road, driving speed, number of turns, etc., to ensure that the calculated path is both efficient and practical. At the same time, the system will also dynamically adjust the path according to the real-time traffic condition information to avoid congestion and delays. In this way, the present invention can provide accurate and reliable navigation services for users, ensuring that users can reach their destinations easily and quickly.

[0030] S203. Display the optimal path as a navigation route to the user and update the traffic condition information in real time.

[0031] It should be noted that in step S203, the system will update the traffic condition information in real time to ensure the accuracy and reliability of the navigation route. Specifically, the system will regularly obtain the latest road conditions, congestion situations, traffic accidents, etc. from traffic data sources and integrate this information into the navigation route. When it detects a change in the road condition, such as congestion or a traffic accident ahead, the system will immediately recalculate the path and provide a new navigation route for the user. In this way, the present invention can provide real-time and dynamic navigation services for users, helping users avoid congestion and delays and ensuring a smooth and efficient journey. In addition, the system will also provide personalized navigation services according to the user's preferences and needs, such as recommending scenic spots, restaurants along the way, etc., to enhance the user's travel experience.

[0032] As an optional embodiment of the present invention, optionally, the remote control of the vehicle in step S3 includes: S301. Based on the vehicle control intention of the user, establish a communication connection with the vehicle through the vehicle control interface; It should be noted that in step S301, the system will first analyze the vehicle control intention of the user, which is usually achieved through natural language processing technology. Once the user's intention is accurately recognized, such as turning on the air conditioner, adjusting the volume or starting the vehicle, etc., the system will immediately establish a communication connection with the vehicle through the vehicle control interface. This interface is designed based on vehicle networking technology, which allows the intelligent assistant to exchange data with the vehicle safely and efficiently. After establishing the connection, the system will send corresponding control instructions to the vehicle according to the user's intention. These instructions will be encrypted to ensure the security of data transmission. After receiving the instructions, the vehicle will immediately execute the corresponding operations and feedback the execution results to the intelligent assistant. In this way, the present invention can achieve the remote control of the vehicle and provide a more convenient and intelligent travel experience for users.

[0033] S302. Generate a remote control instruction according to the user's vehicle control intention, and remotely control the vehicle based on the remote control instruction; It should be noted that in step S302, the system will generate specific remote control instructions according to the user's vehicle control intention. These instructions are used to trigger the vehicle to perform corresponding operations. For example, if the user wishes to turn on the vehicle's air conditioner, the system will generate a data packet containing the air conditioner turn-on instruction and send it to the vehicle through the vehicle control interface. After receiving the instruction, the vehicle will immediately start the air conditioning system and adjust the temperature and wind speed according to the user's settings. Similarly, if the user wishes to adjust the volume of the vehicle or start the vehicle, etc., the system will also generate corresponding remote control instructions and remotely control the vehicle. In this way, the present invention can achieve comprehensive remote control of the vehicle and provide a more intelligent and convenient travel experience for users.

[0034] S303. Real-time feedback the execution result of the remote control to the user.

[0035] It should be noted that in step S303, the system will real-time feedback the execution result of the remote control to the user to ensure that the user can timely understand and control the state of the vehicle. Specifically, once the vehicle executes a remote control instruction, such as turning on the air conditioner or adjusting the volume, etc., the system will immediately receive the data of the execution result from the vehicle. These data will be parsed and processed, and then presented to the user in an intuitive way, such as through the prompt information on the APP interface or voice broadcast, etc. In this way, the user can always master the state and control situation of the vehicle, so as to travel more safely and conveniently. In addition, the system will also record the historical records of the remote control for the user to view and review at any time. These records include information such as the time of control, the executed instruction, and the state of the vehicle, providing comprehensive travel management and monitoring services for the user.

[0036] As an optional embodiment of the present invention, optionally, the method further includes: S6. Record the historical navigation demand information, and interact with the user based on the historical navigation demand information and the current navigation demand information.

[0037] It should be noted that in step S6, the system records the user input (including voice and text input methods) As an optional embodiment of the present invention, optionally, in step S4, when the recognized user intention is an information query intention, according to the query content, retrieving relevant information from the preset vehicle information library and presenting it to the user includes: S401. Based on the user's query content, retrieve vehicle information entries related to the user's query content from the preset vehicle information library; It should be noted that in step S401, the system will use efficient retrieval algorithms, such as inverted index or semantic matching, etc., to quickly locate vehicle information entries related to the user's query content. These entries include the detailed configuration, performance parameters, maintenance records, repair history, etc. of the vehicle. By precisely matching the user's query requirements, the system can provide accurate and comprehensive vehicle information for the user. At the same time, the system will also perform intelligent sorting on the retrieval results according to the user's query history and preferences, and display the information that best meets the user's needs to the user first. In this way, the user can not only quickly obtain the required information, but also obtain a more personalized and considerate service experience.

[0038] S402. Screen and sort the retrieved vehicle information entries, and determine the order to be displayed to the user according to the relevance and importance of the information to the query content.

[0039] It should be noted that in step S402, the system will adopt sorting algorithms, such as Learning to Rank (LTR) or sorting algorithms based on content similarity, to perform intelligent sorting on the retrieved vehicle information entries. These algorithms will comprehensively consider multiple factors such as information relevance, timeliness, and user preferences to ensure that the information that best meets the user's needs is displayed to the user first. In this way, the user can quickly find the required information, reduce the search time, and improve the query efficiency. At the same time, the system will also continuously optimize the sorting algorithm according to the user's feedback and behavior data to improve the intelligence and personalization level of the service.

[0040] Embodiment 2 As Figures 2 to 6 shown, the intelligent assistant navigation system of the car company APP based on the large model, the system includes the above-mentioned intelligent assistant navigation method of the car company APP based on the large model; the system also includes: A user interaction module, which is used to receive the user's navigation demand information, display the navigation route, remote control results, and information query results; As Figure 6 shown, the user interaction module is used to receive the user's voice or text input and convert it into navigation demand information recognizable by the system. This module supports multiple input methods, such as voice commands, text input, or gesture operations, etc., to meet the needs and preferences of different users. Once receiving the user's navigation demand, the user interaction module will immediately pass it to the natural language processing module for processing. At the same time, the user interaction module is also responsible for displaying the navigation route, remote control results, information query results, etc., providing intuitive and clear travel guidance and services for the user. Through close cooperation with the intelligent assistant module, the user interaction module can achieve comprehensive control and management of the vehicle and the navigation system, providing a more convenient and intelligent travel experience for the user.

[0041] A speech recognition module, connected to the user interaction module, for converting voice information input by the user into text information; As Figure 6 shown, the speech recognition module is used to collect the user's voice information and convert it into text information through advanced speech recognition technology. During the speech recognition process, the module will use deep learning algorithms to extract features and match patterns from the speech signal, so as to achieve accurate recognition of the user's instructions. Once the recognition is completed, the speech recognition module will transfer the converted text information to the natural language processing module for further processing and analysis.

[0042] A natural language processing module, connected to the speech recognition module, for identifying the user's intention using the text information; As Figure 6 shown, the natural language processing module identifies the user's intention by receiving the text information input by the user and the text information converted by the speech recognition module. This module uses advanced natural language processing technology to perform semantic analysis and intention recognition on the text information. It can understand the user's complex instructions and requirements, such as navigating to a specific location, querying vehicle information, etc., and convert them into operation instructions that the system can execute. Through the intelligent processing of the natural language processing module, the system can interact with the user more naturally and smoothly, improving the user's travel experience.

[0043] A map data module, connected to the natural language processing module, for storing and providing map data; As Figure 6 shown, the map data module stores a large amount of map information, including key data such as road networks, traffic signs, and geographical locations. These data are the basis for the navigation system to achieve accurate navigation and route planning. By connecting to the natural language processing module, the map data module can quickly provide relevant map information according to the user's navigation needs, ensuring the accuracy and reliability of the navigation route. At the same time, the map data module will also update the map information regularly to reflect the latest road changes and traffic conditions, providing users with more real-time and accurate navigation services.

[0044] A vehicle control interface module, connected to the natural language processing module, for communicating with the vehicle; As Figure 6As shown in the figure, the vehicle control interface module is responsible for implementing the communication and data exchange between the intelligent assistant and the vehicle. It is designed based on vehicle networking technology and has the ability to transmit data efficiently and securely. By connecting to the vehicle control unit, the vehicle control interface module can receive remote control instructions from the intelligent assistant and send them to the vehicle for execution. At the same time, it can also obtain the vehicle's status information in real time, such as vehicle speed, fuel level, door status, etc., and feedback this information to the intelligent assistant so that users can understand and control the vehicle's status at any time. In this way, the vehicle control interface module realizes the seamless connection between the intelligent assistant and the vehicle, providing users with a more convenient and intelligent travel experience.

[0045] The vehicle information database module, connected to the natural language processing module, is used to store vehicle-related information.

[0046] As Figure 6 shown in the figure, the vehicle information database module stores a large amount of vehicle-related information, including the vehicle's detailed configuration, performance parameters, maintenance records, repair history, and owner's manual, etc. These information are important bases for users to understand and control the vehicle's status. By connecting to the natural language processing module, the vehicle information database module can quickly retrieve and provide relevant vehicle information according to the user's query needs. For example, when a user hopes to know the vehicle's maintenance cycle or repair record, the system will parse the user's query intention through the natural language processing module and retrieve the corresponding information from the vehicle information database module to display to the user. In this way, users can not only easily obtain the required information, but also make more informed decisions based on this information, such as arranging vehicle maintenance or repair in a timely manner to ensure the safety and reliability of the vehicle. In addition, the vehicle information database module will regularly update the vehicle information to reflect the latest vehicle configuration and performance changes, providing users with a more comprehensive and accurate vehicle information service.

[0047] The present invention aims at the insufficiently intelligent function navigation operation of current car company APPs. The present invention provides an intelligent assistant navigation method and system based on a large model to optimize the function navigation mode of car company APPs.

[0048] Therefore, the system of the present invention is divided into the following modules: 1. The user interaction module, which receives user input (voice or text) and displays the output content of the intelligent assistant.

[0049] 2. The speech recognition module, which converts speech input into text. Currently, it supports English and Chinese input. This module calls a third-party API for speech-to-text conversion. After the API content is expanded, the present invention will also support it. At the same time, if the user inputs text, this module does not need to function.

[0050] 3. Natural language processing module, which performs intent recognition, named entity recognition, context management, etc. based on a large model (which can be an API that calls a large model or a locally deployed large model).

[0051] 4. Interact with the back-end system of the car company's APP to call functions such as remote control and service query.

[0052] The method of intelligent assistant navigation can be divided into the following four steps: Step 1, user input parsing. The user can complete the input of instructions by voice or text input. If it is voice, a wake-up word can be set. For example: "What is the current status of my tires?", "Help me make an appointment for the next maintenance.", "Where is the nearest charging station?". The voice input is converted into text by the voice recognition module and then passed to the natural language processing module for processing. In the parsing of user input in the present invention, since each car company has some proprietary nouns, different entries of each car company need to be added to the stop words during the processing to ensure the accuracy of user input parsing.

[0053] Step 2, intent recognition and semantic parsing, parsing the user's semantics and recognizing the intent. Based on the large language model, the core intent of the user is parsed. In the present invention, the core intent is divided into two types. One is the request for third-party services. For example: My car is out of gas, and the intent is to find a gas station. The other is the service of the car company itself. For example: Make an appointment for maintenance at 10 o'clock tomorrow, and the intent is service reservation. For text information, key entity information such as time, location, and vehicle status also needs to be extracted. In the above examples, the time entity is 10 o'clock tomorrow, and the operation entity is maintenance reservation. At the same time, according to the capabilities of the large language model, the present invention will also record the historical conversations after the user wakes up, enabling it to understand continuous conversations. For example, the user asks: "Is the pressure of my left front tire normal?" The system replies: "The pressure of your left front tire is within the normal range." Immediately afterwards, the user continues to ask "Does the tire need to be replaced?" At this time, the system will know that the entity we are discussing is the left front wheel, and the system will suggest "The wear condition of your left front wheel is good, and it is not recommended to replace it." Step 3, system feedback and function execution. When the intent recognition is completed, the present invention will call the API interface to execute relevant functions. Here, the present invention is also divided into two categories. One is those that do not require page switching, such as vehicle status query, remote control, etc. For querying tire pressure, battery power, etc., only give the numerical value directly; for remote control, send relevant instructions to the TSP platform, such as turning on the air conditioner, and only give one feedback. The other is those that require switching to other pages, such as navigating to a charging station and opening the built-in navigation page of the app. This part is a solution for knowledge graphing the functions of the app. Which operations each page corresponds to are all inferred in the knowledge graph.

[0054] Step 4: User feedback collection and self - learning optimization. Record user click, search, and usage data, and optimize the recommendation logic in combination with reinforcement learning. And based on the above, continuously train the large language model and update the knowledge graph to improve the semantic understanding ability and the reasoning ability of the knowledge graph. For example, if multiple users select a certain store in the "Navigate to 4S store" function, increase the recommendation weight of that store. Optimize the answer quality of the intelligent assistant of the present invention through user feedback.

[0055] The navigation method of the intelligent assistant of the present invention provides a semantic parsing, intelligent recommendation, and dialogue interaction solution based on a large model, enabling the car company APP to have a more intelligent, efficient, and personalized navigation experience. Users can efficiently access the required services without manually searching for functions, and can improve user satisfaction by simply inputting voice or text. It can effectively enhance the user experience of the car company APP, improve the function access efficiency, and reduce the user operation cost.

[0056] Regarding the improvement of function access efficiency, the experimental data is as follows:

[0057] The intelligent assistant can on average reduce the function access time by 65% - 70%, significantly improving the operation efficiency.

[0058] Figure 6 The overall architecture of the intelligent assistant navigation system of the present invention is shown, including four core modules: 1. User interaction module: Users input voice or text through the APP and obtain feedback from the intelligent assistant. 2. Speech recognition module: Convert voice input into text, supporting Mandarin, English, and some dialects. 3. Natural Language Processing (NLP) module: Based on a large model (local or API call), parse user intentions, named entities, and manage the dialogue context. 4. Car company APP background interaction module: The intelligent assistant docks with the car company's back - end system through the API to achieve remote control (such as unlocking, starting) or query services (such as battery power, maintenance appointment).

[0059] Figure 2 Describes the interaction process between the user and the intelligent assistant on the APP: 1. The user inputs instructions in voice or text mode.

[0060] 2. If it is voice input, the speech recognition module converts it into text; if it is text input, it directly enters the next step.

[0061] 3. The NLP module performs intention recognition and entity extraction on the user input to understand the user's needs.

[0062] 4. If the user queries vehicle information or control functions, the system calls the car company API to obtain relevant data or perform operations.

[0063] 5. The vehicle enterprise's back-end returns data, which is sorted out by the intelligent assistant and fed back to the user in the form of text or voice.

[0064] Figure 3 The working process of the speech recognition module is shown: 1. The user inputs voice (such as "Help me turn on the air conditioner").

[0065] 2. The voice data is optimized through noise reduction and signal processing to improve the recognition accuracy.

[0066] 3. An ASR (Automatic Speech Recognition) model is adopted to convert the voice into text.

[0067] 4. Post-processing is performed on the recognition result (such as accent adaptation and typo correction).

[0068] 5. If the recognition is successful, the text input is passed to the NLP module; if the recognition fails, the user is prompted to re-enter.

[0069] Figure 4 Describe in detail how the NLP module processes the user input: 1. Intent recognition: Distinguish whether the user is querying vehicle information, controlling the vehicle, or seeking help.

[0070] 2. Named Entity Recognition (NER): Extract key information, such as license plate number, destination, time, etc.

[0071] 3. Context management: Ensure the continuity of multi-round conversations. For example: User: "How much power does the car have left now?" Assistant: "The current power is 78%." User: "Where is the nearest charging station?" (The assistant needs to remember the context related to "power") 4. The processed data is handed over to the API call module to request the vehicle enterprise's back-end to execute operations or query information.

[0072] 5. The result is returned and shows how the intelligent assistant interacts with the vehicle enterprise's back-end system:

[0073] Figure 5 Shows how the intelligent assistant interacts with the vehicle enterprise's back-end system: 1. The NLP parses the user request, such as "Help me unlock the car door" or "Query the cruising range";

[0074] 2. Select the appropriate API endpoint: Remote control (such as / vehicle / unlock) Data query (such as / vehicle / battery) Schedule service (e.g. / service / schedule); 3. Attach OAuth2 authentication token when sending API requests to ensure secure access;

[0075] 4. The car company server processes the request and returns data in JSON format;

[0076] 5. The data is analyzed by the intelligent assistant and fed back to the user in the form of text or voice.

[0077] The embodiment of the present invention describes an intelligent assistant navigation system, which can be embedded in the car company's APP, and realizes intelligent navigation and vehicle control functions through voice recognition, natural language processing (NLP) and remote background interaction. Figure 6 As shown, the overall process includes the following steps: 1. The user initiates a request (voice or text) through the APP.

[0078] 2. The voice input is converted into text through the voice recognition module (if it is text input, skip this step).

[0079] 3. The NLP module analyzes the user's intention, determines whether the user is querying information or controlling vehicle functions, and extracts relevant information (such as destination, vehicle condition, time, etc.).

[0080] 4. NLP calls the car manufacturer's API to perform corresponding operations (such as checking the battery level, unlocking the door, turning on the air conditioner, etc.).

[0081] 5. The execution results are returned to the user and displayed interactively through the smart assistant interface.

[0082] Example: Remote unlocking operation: User: "Help me unlock the door."; Voice recognition module converts to text: "Help me unlock the door."; NLP analysis: Intent recognition (unlocking), object (vehicle); API request: POST / vehicle / unlock; Car company backend response: "Unlocking successful."; Smart assistant feedback: "The door is unlocked." Improvements to the natural language processing (NLP) module. This example describes in detail Figure 5 The natural language processing module in the paper focuses on how to improve the understanding ability of the intelligent assistant by calling the large model API or locally deploying the large model (the method adopted by the present invention is to call the DeepSeek API).

[0083] 1. Intent Recognition: Use prompt engineering of large models to recognize the intent of the text. In the present invention, the functions of the car company APP are classified through analysis, including querying vehicle status; service reservation; location query; and remote control. Prompt engineering optimization is carried out for each of them respectively. Of course, the functions of different car company APPs will be different, and different prompt engineering methods can be adopted according to different APPs in the present invention. Such as "query vehicle status" or "execute vehicle control operation".

[0084] 2. Named Entity Recognition (NER): Extract key entities from the user input, such as license plate number, destination, power information, etc.; In the present invention, entity recognition is carried out by analyzing the entity list that can be included in the car company APP and combining the capabilities of large models. Specifically, sequence labeling can be achieved by adding a layer specific to the NER task in the output layer, and at the same time, prompt engineering is combined to improve the accuracy of entity recognition. And there are different professional terms according to different car companies during entity recognition, which also improves the accuracy of entity recognition to a certain extent.

[0085] 3. Context Management: Build the user conversation history to ensure the consistency of multi-round conversations. For example, optimization of multi-round conversation experience: User: "How much power does my car have left?"; NLP parsing: Intent = query power; API call: GET / vehicle / battery; The car company backend returns: "The current power is 78%."; User: "Can you help me find the nearest charging station?"; NLP parsing: "Nearest" refers to the user's current location, and "charging station" belongs to the category of charging facilities; API call: GET / charging_stations?location=current; The car company backend returns: "The nearest one is XX charging station, 3.2 km away."; The intelligent assistant feedback: "The nearest charging station is XX, 3.2 km away. Do you need navigation?" This improvement plan improves the accuracy of natural language understanding and enhances the interaction experience, enabling the assistant to remember the user context and not requiring the user to re-enter complete information every time.

[0086] For Figure 5 the interaction module with the car company backend, the embodiments of the present invention focus on describing intent recognition and calling the car company backend API. 1. Through the natural language processing module, when the intent recognition is to directly call the API provided by the car company; 2. Through the natural language processing module, named entity recognition identifies specific entities, mainly for the parameters when calling the API. For example, User: "Query the tire pressure of the left front tire." Through the natural language processing module, it is known that the intent is to query the tire pressure; and the parameter is the left front wheel. 3. Call the API. In order to achieve secure communication in the present invention, all API requests need to carry the user identity token (OAuth2) to prevent unauthorized access. The present invention uses data encryption to ensure vehicle control commands.

[0087] Testing and optimizing the navigation function of the intelligent assistant. Embodiments of the present invention are based on Figure 2 , test the interaction process of the intelligent assistant to ensure that the system can operate normally under different application scenarios.

[0088] Test environment: Device: mobile phone; Network environment: 4G / 5G / WiFi; Test API response time; Test NLP parsing accuracy; The test results are as follows:

[0089] Through the detailed description of the above multiple embodiments, the intelligent assistant navigation system of the present invention can efficiently parse the user's intention, seamlessly connect with the background of the car company APP, and provide accurate vehicle remote control and service query functions. Compared with the prior art, the present invention has the following core advantages: 1. Intelligent voice interaction: Provide a natural conversational interaction experience by combining ASR and NLP. 2. Accurate intention recognition: Optimize based on the large model to improve the user's intention understanding ability. 3. Efficient remote control: Optimize API calls to improve the vehicle control execution speed. 4. Enhanced security: Support OAuth2 identity authentication to prevent unauthorized access.

[0090] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and purposes of the present invention. The scope of the present invention is defined by the claims and their equivalents.

Claims

1. A large model-based car enterprise APP intelligent assistant navigation method, characterized in that: The method comprises: S1. Collecting navigation demand information of the user, obtaining key entities based on the navigation demand information using a large language model, and identifying the user's intention based on the key entity information; the expression for obtaining the key entity using the large language model is: ； in, Entities represents the final extracted entity set, ( w l * , tag l * ) represents an entity pair, consisting of a word w l * and its tags tag l * composition, Indicates that in all label sequences tag In , find the label sequence that maximizes the probability product, P ( tag l |w * , tag 1:l-1 ) means that in a given text w * and History tab tag 1:l-1 Under the conditions of l Entities belong to the tag tag l The probability of tag 1:l-1 Indicates that from l Entity to l -1 entity label sequence; S2. When the user intention is identified as navigation intention, a navigation route is generated according to the current location and the target location using map data and a path planning algorithm; S3. When the identified user intention is a vehicle control intention, based on the user's vehicle control intention, communicate with the vehicle through a vehicle control interface to remotely control the vehicle; S4. When the user intention is identified as information query intention, relevant information is retrieved from a preset vehicle information database according to the query content and displayed to the user; S5. Collect user feedback, and optimize the large language model and navigation strategy based on the feedback.

2. The large model-based car enterprise APP intelligent assistant navigation method according to claim 1 is characterized in that: In step S1, the navigation requirement information includes text information and voice information; When the user inputs voice information, a wake-up word is set, and the voice information is converted into text information using a speech recognition method. The large language model is then used to extract key entities and identify user intent.

3. The large model-based car enterprise APP intelligent assistant navigation method as claimed in claim 2 is characterized in that: The expression for converting the voice information into text information is: ， ， ， ， ， ， ； in, y(t) represents a noisy speech signal, t represents the time variable, x(t) represents the original speech signal, n(t) represents the environmental noise signal, y filt (t) represents the enhanced speech signal, * represents the convolution operator, h(t) represents the impulse response function, y(τ) Represents the original noisy speech signal at time τ The amplitude of h(t-τ) It represents the impulse response function in time difference t-τ The gain when Y m,k Indicates m Frame, k The complex spectral coefficients of the frequency points, N Indicates the frame length. n Indicates the frame number. m Represents the frame index, H represents frame shift, w(n) represents the window function, e -j2πkn / N represents the complex twiddle factor, MFCC m Indicates m Mel-frequency cepstral coefficient vector of the frame, DCT ( ) represents discrete cosine transform, b i Indicates i Mel filter band boundaries, M i Indicates i Mel filter weights, h m represents the encoded hidden state vector, Encoder ( ) indicates the encoder, P (w) represents the joint probability of the output text sequence w, L Represents the length of the output text sequence, w l Indicates l word, Represents the final optimized text output, P ASR (w) represents the word sequence probability output by the speech recognition model, λ represents the language model weight coefficient, P LM (w) represents the probability of grammatical correction given by the language model.

4. The large model-based car enterprise APP intelligent assistant navigation method according to claim 1, characterized in that: In step S1, the expression for identifying the user intention based on the key entity information is: ； in, I * represents the optimal user intention, Indicates that in all intent categories C Find the category that maximizes the probability c k , C represents the set of all intent categories, Intent category c k The corresponding weight vector, L It represents the length of the input entity sequence. represents the aggregated entity feature vector, e l Indicates l The word embedding vector of each entity, t l Indicates l The label embedding vector of each entity, represents the vector concatenation operation, Intent category c k The corresponding bias term is, K Indicates the number of categories for the classification task.

5. The large model-based car enterprise APP intelligent assistant navigation method according to claim 1, characterized in that: Generating a navigation route using map data and a path planning algorithm in step S2 includes: S201, based on the current position and the target position, obtaining corresponding road network information and traffic condition information from map data; S202, using a path planning algorithm to calculate an optimal path from the current location to the target location based on the road network information and traffic condition information; S203: Display the optimal path as a navigation route to the user, and update traffic status information in real time.

6. The large model-based car enterprise APP intelligent assistant navigation method according to claim 1, characterized in that: In step S3, remotely controlling the vehicle includes: S301, establishing a communication connection with the vehicle through a vehicle control interface based on the vehicle control intention of the user; S302, generating a remote control instruction according to the vehicle control intention of the user, and remotely controlling the vehicle based on the remote control instruction; S303: Feedback the execution result of the remote control to the user in real time.

7. The large model-based car enterprise APP intelligent assistant navigation method according to claim 1, characterized in that: The method further comprises: S6. Record historical navigation demand information, and interact with the user based on the historical navigation demand information and current navigation demand information.

8. The large model-based car enterprise APP intelligent assistant navigation method according to claim 1, characterized in that: In step S4, when the user intention is identified as information query intention, retrieving relevant information from a preset vehicle information database and displaying it to the user according to the query content includes: S401, based on the query content of the user, retrieving vehicle information items related to the query content of the user from a preset vehicle information database; S402: Filter and sort the retrieved vehicle information items, and determine the order in which the information is displayed to the user based on the relevance and importance of the information to the query content.

9. The car enterprise APP intelligent assistant navigation system based on the big model is characterized by: The system adopts the large model-based car enterprise APP intelligent assistant navigation method according to any one of claims 1 to 8; the system also includes: A user interaction module is used to receive navigation demand information from users, display navigation routes, remote control results and information query results; A speech recognition module, connected to the user interaction module, for converting the speech information input by the user into text information; A natural language processing module, connected to the speech recognition module, for identifying user intentions using the text information; A map data module, connected to the natural language processing module, for storing and providing map data; A vehicle control interface module, connected to the natural language processing module, for communicating with the vehicle; The vehicle information database module is connected to the natural language processing module and is used to store vehicle related information.

Images