Vehicle scene arrangement recommendation method and device, equipment and automobile

Abstract

The embodiment of the invention provides a vehicle scene arrangement recommendation method and device, equipment and a vehicle. The vehicle scene arrangement recommendation method comprises the steps of obtaining a current user demand of a first vehicle; obtaining first historical data, wherein the first historical data comprises a user historical demand and a corresponding first control domain chain; determining a first arrangement result for vehicle scene arrangement by using a first recommendation model obtained by pre-training and the current demand of the user; the first arrangement result comprises a second control domain chain; and selecting a first control domain chain matched with the current demand of the user from the first historical data as a second arrangement result for vehicle scene arrangement, and recommending the first arrangement result and the second arrangement result.

Description

Technical Field

[0001] The present application relates to the field of automobile technology, and in particular to a vehicle scene arrangement recommendation method and device, equipment, and automobile. Background Art

[0002] The scene arrangement service of the smart cockpit can realize the scene-based customized services of the vehicle, such as automatic adjustment of the vehicle temperature, timed automatic navigation, automatic control of the body electronics, personalized rest mode, etc., which can enhance the user experience and meet the user's personalized service needs.

[0003] However, the scene arrangement service is difficult to use in real vehicles, especially for users who are not familiar with the function, and the service function cannot play its practical role. Summary of the invention

[0004] In a first aspect, an embodiment of the present application provides a vehicle scene arrangement recommendation method, the method comprising: obtaining a user's current demand for a first vehicle; and obtaining first historical data, the first historical data comprising the user's historical demand and a corresponding first control domain chain; and determining a first arrangement result for vehicle scene arrangement using a pre-trained first recommendation model and the user's current demand; the first arrangement result comprising a second control domain chain; and selecting a first control domain chain matching the user's current demand from the first historical data as a second arrangement result for vehicle scene arrangement; and recommending the first arrangement result and the second arrangement result.

[0005] It can be understood that in the embodiment of the present application, for the vehicle scene orchestration service, not only is the first orchestration result that meets the user's current needs recommended to the user based on the pre-trained first recommendation model, but the second orchestration result that matches the user's current needs is also selected based on the user's historical needs and the corresponding first control domain chain (i.e., the first historical data); based on this, both the first orchestration result and the second orchestration result are recommended to the user, thereby solving the difficulty in using the scene orchestration service in actual vehicles and lowering the threshold for using the vehicle scene orchestration service, while giving users more choices and ensuring that users can select a more satisfactory vehicle scene orchestration result from multiple orchestration results.

[0006] Further, in some embodiments, obtaining the current needs of the user of the first vehicle includes: obtaining voice data for the first vehicle, the voice data being used to express the current needs of the user; determining a first arrangement result for vehicle scene arrangement using a pre-trained first recommendation model and the current needs of the user, including: performing voice feature preprocessing on the voice data to obtain voice features; and inputting the voice features into a voice-word vector model to obtain a word embedding vector; and determining the first arrangement result using the pre-trained first recommendation model and the word embedding vector.

[0007] It can be understood that in the embodiment of the present application, after obtaining the voice data used to express the user's current needs, the voice data is pre-processed with voice features, and then the obtained voice features are directly input into the voice-word vector model, rather than translating the obtained voice features into text data and then inputting them into the artificial intelligence (AI) model; in this way, the processing link between the voice data and the first arrangement result is shortened, and since the process of translating the voice features into text data through the voice feature conversion text module is eliminated, it is beneficial to reduce semantic errors in the voice conversion process and reduce the misrecognition rate, which is beneficial to obtaining a more accurate first arrangement result, and further beneficial to optimizing the business performance of the vehicle scene arrangement service.

[0008] Further, in some embodiments, the determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector without querying the LORA parameters of the first recommendation model.

[0009] It can be understood that in the embodiment of the present application, it is first queried whether the LORA parameters of the first recommendation model exist. If they do not exist, that is, the LORA parameters of the first recommendation model are not queried, it means that the first recommendation model cannot be used to recommend personalized vehicle scene arrangement results for users. At this time, the pre-trained first recommendation model and the word embedding vector are still used to recommend vehicle scene arrangement results (i.e., the first arrangement results) to users, thereby lowering the usage threshold for new users and improving the popularity and usability of vehicle scene arrangement service functions.

[0010] Further, in some embodiments, the determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: when the LORA parameters of the first recommendation model are queried, determining the first arrangement result by using the pre-trained first recommendation model, the LORA parameters and the word embedding vector.

[0011] It can be understood that in the embodiment of the present application, it is first queried whether the LORA parameters of the first recommendation model exist. If they exist, that is, the LORA parameters of the first recommendation model are queried, the pre-trained first recommendation model, the LORA parameters and the word embedding vector are used to recommend the vehicle scene arrangement result (that is, the first arrangement result) to the user, so that the first arrangement result can better meet the user's personalized customization needs and improve the vehicle's intelligence level and user satisfaction.

[0012] In some embodiments, the obtaining of the first historical data includes: when no historical tracking data of the first vehicle is found, using pre-selected recommended special data as the first historical data; the recommended special data includes user historical demands of multiple different vehicles and the corresponding first control domain chain.

[0013] It can be understood that in the embodiment of the present application, it is first queried whether the historical burial point data of the first vehicle exists. If not, the pre-selected recommended special data is used as the first historical data, and the second arrangement result matching the current needs of the user is selected from the recommended special data and recommended to the user; in this way, the second arrangement result recommended to the user is more in line with the current needs of the user, thereby helping the user to quickly use the vehicle scene arrangement service.

[0014] In some embodiments, the obtaining of the first historical data includes: when historical burial point data of the first vehicle is queried, using the historical burial point data as the first historical data.

[0015] It can be understood that in the embodiment of the present application, it is first queried whether the historical tracking data of the first vehicle exists. If so, the historical tracking data is used as the first historical data, and based on this, a second scheduling result matching the current needs of the user is selected from the historical tracking data and recommended to the user. In this way, the obtained second scheduling result is more in line with user preferences, thereby improving the intelligence level and user satisfaction of the vehicle scene scheduling service.

[0016] Furthermore, in some embodiments, the method further includes: in the case where the LORA parameters of the first recommendation model are not queried, performing LORA fine-tuning on the first recommendation model according to the historical burial point data to obtain the LORA parameters of the first recommendation model.

[0017] It can be understood that in an embodiment of the present application, when the LORA parameters of the first recommendation model are not queried but the historical tracking data exist, the first recommendation model is fine-tuned by LORA according to the historical tracking data to obtain the LORA parameters of the first recommendation model; in this way, in the subsequent vehicle scene orchestration recommendations, the LORA parameters can be used and combined with the first recommendation model to perform personalized recommendations for vehicle scene orchestration, so that the recommended orchestration results can better meet the user's private customization needs, thereby improving the intelligence level of the vehicle scene orchestration service and user satisfaction.

[0018] In a second aspect, an embodiment of the present application provides a vehicle scene arrangement recommendation device, the vehicle scene arrangement recommendation device comprising: a first acquisition module, configured to obtain the user's current demand for a first vehicle; a second acquisition module, configured to obtain first historical data, the first historical data including the user's historical demand and the corresponding first control domain chain; a first determination module, configured to use a pre-trained first recommendation model and the user's current demand to determine a first arrangement result for the vehicle scene arrangement; the first arrangement result includes a second control domain chain; a second determination module, configured to select a first control domain chain that matches the user's current demand from the first historical data as a second arrangement result for the vehicle scene arrangement; a recommendation module, configured to recommend the first arrangement result and the second arrangement result.

[0019] In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can be executed on the processor, and when the processor executes the program, the vehicle scene arrangement recommendation method described in the first aspect is implemented.

[0020] In a fourth aspect, an embodiment of the present application provides a car, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and when the processor executes the program, the vehicle scene arrangement recommendation method described in the first aspect is implemented.

[0021] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the method described in the first aspect when executed by a processor, an electronic device, or a car.

[0022] In a sixth aspect, an embodiment of the present application provides a computer program product, including a computer program or instructions, which, when executed by a processor, an electronic device, or a car, implements the method described in the first aspect of the present application.

[0023] In a seventh aspect, an embodiment of the present application provides a computer program, which enables a processor, an electronic device, or a car to execute the method described in the first aspect.

[0024] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solution of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0026] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0027] Figure 1 A schematic diagram of the implementation process of the vehicle scene arrangement recommendation method provided in an embodiment of the present application;

[0028] Figure 2 A schematic diagram showing a process comparison between the vehicle scenario arrangement recommendation method provided in an embodiment of the present application and a general method;

[0029] Figure 3 A schematic diagram of the steps for training and fine-tuning the general recommendation model provided in an embodiment of the present application;

[0030] Figure 4 A schematic diagram of a specific usage flow of a general recommendation model provided in an embodiment of the present application;

[0031] Figure 5 A schematic diagram of the structure of a vehicle scene arrangement recommendation device provided in an embodiment of the present application;

[0032] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the specific technical solution of the present application will be further described in detail below in conjunction with the drawings in the embodiments of the present application. The following embodiments are used to illustrate the present application, but are not used to limit the scope of the present application.

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0035] In the following description, reference is made to “some embodiments”, “this embodiment”, “embodiments of the present application” and examples, etc., which describe a subset of all possible embodiments, but it can be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0036] The descriptions of “first, second, third” etc. that appear in the embodiments of the present application are only for illustration and distinction of the description objects. There is no distinction of order, nor do they indicate any special limitation on the number of devices in the embodiments of the present application, and cannot constitute any limitation on the embodiments of the present application.

[0037] The “big business model” and “general recommendation model” that appear in the embodiments of the present application can be understood as the “first recommendation model obtained by pre-training”.

[0038] To facilitate understanding of the technical solutions of the embodiments of the present application, the following describes the related technologies or terms of the embodiments of the present application. The following related technologies or related terms can be combined arbitrarily with the technical solutions of the embodiments of the present application as optional solutions, and they all belong to the protection scope of the embodiments of the present application.

[0039] (1) Vehicle scenario arrangement service

[0040] Vehicle scenario orchestration service, also known as vehicle-side scenario orchestration service, allows users to customize various vehicle functions according to their needs and preferences, thereby creating personalized driving scenarios.

[0041] For example, users can customize one or more of the following functions of the vehicle: driving assistance function, infotainment function, smart connectivity function, safety protection function, comfort and convenience function; among which:

[0042] Driving assistance functions include but are not limited to one or more of the following: automatic parking, adaptive cruise control, lane keeping;

[0043] Infotainment functions include but are not limited to one or more of the following: in-car navigation, music playback, video playback;

[0044] Smart interconnection functions include but are not limited to one or more of the following: remote control, vehicle status monitoring, and Internet of Vehicles services;

[0045] Safety protection functions include but are not limited to one or more of the following: collision warning, emergency braking, 360-degree panoramic imaging;

[0046] Comfort and convenience features include but are not limited to one or more of the following: automatic air conditioning, automatic windows, automatic sunroof, seat heating / ventilation, and intelligent voice assistant.

[0047] With the continuous improvement of the intelligence level of vehicle-side services, users can enjoy more abundant customized services. At present, the vehicle scene arrangement service can automatically complete the scheduling and execution according to some instructions of the user, which brings great convenience to the user, saves time and worry, and significantly improves the user experience. However, the product often needs to go through a transition stage from initial contact to proficient use. Therefore, recommending the highlights of the product to users in a way that is closer to the user's real use scenario will help users quickly become familiar with and increase the usage rate of the function. In addition, the voice interaction method has improved the convenience and interactivity of vehicle scene arrangement to a certain extent. Although complex voice arrangement functions can be realized with the help of voice-text-large language model technology, the voice-to-text method is limited by many factors such as its accuracy and environmental noise, which will indirectly affect the practicality and convenience of the vehicle scene arrangement service to a certain extent. Therefore, solving the problems existing in the user's use process and the experience problems in the voice-to-text process are of great significance to the promotion and application of vehicle scene arrangement services.

[0048] The inventors of this application discovered during the process of collecting and studying the actual usage of vehicle scenario orchestration services:

[0049] 1. The voice processing link of the in-car voice scene arrangement service is too long, and the inaccurate voice recognition leads to unsatisfactory results of the voice arrangement service.

[0050] 2. The vehicle scene arrangement service is difficult to use in real vehicles, especially for users who are not familiar with the function, and the service function cannot play its practical role.

[0051] Based on the above problems, the present application provides multiple embodiments of the following vehicle scenario arrangement recommendation method.

[0052] Figure 1 A schematic diagram of the implementation flow of the vehicle scene arrangement recommendation method provided in an embodiment of the present application; Figure 1 As shown, the method may include the following steps 101 to 105:

[0053] Step 101, obtaining the current demand of the user of the first vehicle;

[0054] Step 102, obtaining first historical data, where the first historical data includes user historical demands and corresponding first control domain chains;

[0055] Step 103, using the pre-trained first recommendation model and the current needs of the user, determining a first arrangement result for vehicle scenario arrangement; the first arrangement result includes a second control domain chain;

[0056] Step 104, selecting a first control domain chain matching the current demand of the user from the first historical data as a second arrangement result for vehicle scenario arrangement;

[0057] Step 105: recommend the first arrangement result and the second arrangement result.

[0058] It can be understood that in the embodiment of the present application, for the vehicle scene orchestration service, not only is the first orchestration result that meets the user's current needs recommended to the user based on the pre-trained first recommendation model, but the second orchestration result that matches the user's current needs is also selected based on the user's historical needs and the corresponding first control domain chain (i.e., the first historical data); based on this, both the first orchestration result and the second orchestration result are recommended to the user, thereby solving the difficulty in using the scene orchestration service in actual vehicles and lowering the threshold for using the vehicle scene orchestration service, while giving users more choices and ensuring that users can select a more satisfactory vehicle scene orchestration result from multiple orchestration results.

[0059] The following describes further optional implementations and related terms of each of the above steps.

[0060] Step 101, obtaining the current demand of the user of the first vehicle.

[0061] In the embodiment of the present application, there is no limitation on the method for expressing the current needs of the user, which may be voice, text, picture or video, etc. The current needs of the user may be actively input by the user, for example, the user inputs "I am hot" or "It is a bit hot in the car" or "I am cold" or "Play some refreshing music" etc. by voice or text.

[0062] Of course, the user's current demand can also be collected through the sensors of the first vehicle without the user's participation, for example, pictures or videos of the environment inside and outside the vehicle collected by the camera on the first vehicle, or the temperature inside the vehicle collected by the temperature sensor on the first vehicle.

[0063] In some embodiments, step 101 includes: acquiring voice data for the first vehicle, wherein the voice data is used to express the current needs of the user.

[0064] Step 102: Acquire first historical data, where the first historical data includes user historical demands and corresponding first control domain chains.

[0065] In the embodiment of the present application, the user's historical demand can be for the first vehicle or for multiple different vehicles. It can be understood that the so-called control domain chain can be understood as the execution information of one or more domains of the vehicle (such as at least one of the power domain, chassis domain, cockpit domain, autonomous driving domain, body domain, etc.), and the control domain chain is decomposed into several domain events, each domain event includes a domain controller and execution information, and the domain controller executes the corresponding execution information.

[0066] In some embodiments, step 102 includes: when no historical tracking data of the first vehicle is found, using pre-selected recommended dedicated data as the first historical data; the recommended dedicated data includes user historical needs of multiple different vehicles and the corresponding first control domain chain.

[0067] In the embodiment of the present application, the historical tracking data of the first vehicle records the actual use information of the first vehicle, and the historical tracking data may include the actual use information of the first vehicle for the most recent N times, where N is greater than 0. It can be understood that the historical tracking data of the first vehicle includes the user's historical needs of the first vehicle and the corresponding first control domain chain.

[0068] In some embodiments, the recommended dedicated data is representative data selected from historical usage information of multiple different vehicles. For example, some reasonable and practical data are selected from the historical compilation data of all users, and then representative data is carefully selected from the selected partial data as the recommended dedicated data.

[0069] It can be understood that in the embodiment of the present application, it is first queried whether the historical burial point data of the first vehicle exists. If not, the pre-selected recommended special data is used as the first historical data, and the second arrangement result matching the current needs of the user is selected from the recommended special data and recommended to the user; in this way, the second arrangement result recommended to the user is more in line with the current needs of the user, thereby helping the user to quickly use the vehicle scene arrangement service.

[0070] In some embodiments, step 102 includes: when historical burial point data of the first vehicle is queried, using the historical burial point data as the first historical data.

[0071] It can be understood that in the embodiment of the present application, it is first queried whether the historical tracking data of the first vehicle exists. If so, the historical tracking data is used as the first historical data, and based on this, a second scheduling result matching the current needs of the user is selected from the historical tracking data and recommended to the user. In this way, the obtained second scheduling result is more in line with user preferences, thereby improving the intelligence level and user satisfaction of the vehicle scene scheduling service.

[0072] Step 103, using the pre-trained first recommendation model and the current needs of the user, determine a first arrangement result for vehicle scenario arrangement; the first arrangement result includes a second control domain chain.

[0073] It should be noted that, for the second control domain chain, the description of the control domain chain can be referred to above for understanding. In addition, in the embodiment of the present application, there is no restriction on the type and structure of the first recommendation model obtained by pre-training. In short, the first recommendation model can obtain a first arrangement result that meets the current needs of the user. For example, the first recommendation model is a large language model, and the large language model can be a llama, glm, baichuan and other models. In one possible implementation, in order to enable the model to understand the user's usage habits, a large language model with relatively small parameters can be selected as the basic model for training. The model has a certain text comprehension ability and can better learn and generalize the annotated data.

[0074] For the first recommendation model, during the training phase, the orchestration history data of the orchestration service backend is first sorted. The orchestration history data is the usage record of all users who will use the vehicle scene orchestration function, such as "[condition: {time, 12:00-1:00; temperature, 25; main driver, someone; humidity, 6%; ...}, action: {air conditioning, 23; car window, closed; ... "}]. All the orchestration history data are clustered and cleaned to remove unreasonable data, leaving a variety of scene data that are similar or identical to most users' orchestrations as training data for the first recommendation model. Based on this, the first recommendation model is trained using the training data. In a possible implementation, the first recommendation model is trained using supervised learning; wherein, the condition of the training data is input into the first recommendation model, and action is used as label information to determine the similarity between the actual output of the first recommendation model and the action (for example, by calculating the similarity of the statement through rules), and back propagation is performed based on the distance between the similarity and the preset threshold. Since this model is a recommendation model, its output does not have a standard answer, but it must ensure that the content is reasonable, the format is correct, and the similarity meets the requirements. After multiple iterations of training, the model's ability is judged based on the accuracy of the first recommendation model on the test set and the validation set. The version with the best model ability is selected as the first recommendation model obtained by pre-training.

[0075] In some embodiments, step 103 includes: performing speech feature preprocessing on the speech data to obtain speech features; and inputting the speech features into a speech-word vector model to obtain a word embedding vector; and determining the first arrangement result using a pre-trained first recommendation model and the word embedding vector.

[0076] It can be understood that the speech features are input into the speech-word vector model, and the model operates on the speech features to obtain / output a word embedding vector.

[0077] It can be understood that in the embodiment of the present application, after obtaining the voice data used to express the user's current needs, the voice data is pre-processed with voice features, and then the obtained voice features are directly input into the voice-word vector model, rather than translating the obtained voice features into text data and then inputting them into the AI ​​model; in this way, the processing link between the voice data and the first arrangement result is shortened, and since the process of translating the voice features into text data through the voice feature conversion text module is eliminated, it is beneficial to reduce semantic errors in the voice conversion process and reduce the misrecognition rate, which is beneficial to obtaining a more accurate first arrangement result, and further beneficial to optimizing the business performance of the vehicle scenario arrangement service.

[0078] Further, in some embodiments, the determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector when the LORA (Low-Rank Adaptation) parameters of the first recommendation model are not queried.

[0079] In some embodiments, the determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: inputting the word embedding vector into the pre-trained first recommendation model, the model performing operations on the word embedding vector to obtain / output the first arrangement result.

[0080] It can be understood that in the embodiment of the present application, it is first queried whether the LORA parameters of the first recommendation model exist. If they do not exist, that is, the LORA parameters of the first recommendation model are not queried, it means that the first recommendation model cannot be used to recommend personalized vehicle scene arrangement results for users. At this time, the pre-trained first recommendation model and the word embedding vector are still used to recommend vehicle scene arrangement results (i.e., the first arrangement results) to users, thereby lowering the usage threshold for new users and improving the popularity and usability of vehicle scene arrangement service functions.

[0081] Further, in some embodiments, the determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: when the LORA parameters of the first recommendation model are queried, determining the first arrangement result by using the pre-trained first recommendation model, the LORA parameters and the word embedding vector.

[0082] It can be understood that in the embodiment of the present application, it is first queried whether the LORA parameters of the first recommendation model exist. If they exist, that is, the LORA parameters of the first recommendation model are queried, the pre-trained first recommendation model, the LORA parameters and the word embedding vector are used to recommend the vehicle scene arrangement result (that is, the first arrangement result) to the user, so that the first arrangement result can better meet the user's personalized customization needs and improve the vehicle's intelligence level and user satisfaction.

[0083] In some embodiments, the first arrangement result is determined by using the pre-trained first recommendation model, the LORA parameters and the word embedding vector, including: inputting the word embedding vector into the pre-trained first recommendation model, the model loading the LORA parameters and performing operations on the word embedding vector to obtain / output the first arrangement result.

[0084] In some embodiments, the method further includes: when the LORA parameters of the first recommendation model are not queried, performing LORA fine-tuning on the first recommendation model according to the historical burial point data to obtain the LORA parameters of the first recommendation model.

[0085] It can be understood that in an embodiment of the present application, when the LORA parameters of the first recommendation model are not queried but the historical tracking data exist, the first recommendation model is fine-tuned by LORA according to the historical tracking data to obtain the LORA parameters of the first recommendation model; in this way, in the subsequent vehicle scene orchestration recommendations, the LORA parameters can be used and combined with the first recommendation model to perform personalized recommendations for vehicle scene orchestration, so that the recommended orchestration results can better meet the user's private customization needs, thereby improving the intelligence level of the vehicle scene orchestration service and user satisfaction.

[0086] In one possible implementation, data can be buried in the vehicle side (first vehicle) in advance to record the user's vehicle-side control habits, extract the service information contained in the vehicle scene arrangement service (that is, historical buried data), and use the vehicle-side service to automatically annotate it as fine-tuning data, such as "[condition: {time, 8:00; temperature, 15; passenger seat, someone; ...}, action: {music, on; air conditioning, 26, ...}]". This type of data includes the user's actual use of the first vehicle, and this type of data is used as the user's personalized data set. Secondly, use the personalized data set to perform LORA fine-tuning on the pre-trained first recommendation model, so that the fine-tuned model has the function of personalized recommendation.

[0087] In one possible implementation, the historical buried data of the first vehicle is detected from the vehicle side. If the historical buried data meets the fine-tuning requirements, the pre-trained first recommendation model is fine-tuned using the historical buried data. During the fine-tuning process, the condition in the historical buried data is input into the pre-trained first recommendation model, and the first recommendation model is fine-tuned according to the user preference (i.e., action) of the first vehicle, so that the action result output by the first recommendation model is close to the corresponding action in the historical buried data (i.e., the user preference of the first vehicle). After the fine-tuning is completed, the LORA parameters for the first vehicle are obtained. When in use, the LORA parameters of the user / the vehicle are loaded into the pre-trained first recommendation model to make personalized recommendations for the user of the first vehicle.

[0088] Step 104: Select a first control domain chain that matches the current demand of the user from the first historical data as a second arrangement result for vehicle scenario arrangement.

[0089] In some embodiments, the selecting a first control domain chain matching the current demand of the user from the first historical data as the second orchestration result for the vehicle scenario orchestration includes: selecting from the first historical data user historical demands whose similarity with the current demand of the user meets similarity conditions; and taking the first control domain chain corresponding to the user historical demands that meet the similarity conditions in the first historical data as the second orchestration result.

[0090] In a possible implementation manner, the similarity condition includes that the similarity with the current demand of the user is greater than or equal to a first threshold.

[0091] Step 105: recommend the first arrangement result and the second arrangement result.

[0092] For example, the first and second arrangement results are displayed on an onboard terminal of the first vehicle so that a user can select a satisfactory arrangement result as a target arrangement result; and the first vehicle executes the target arrangement result.

[0093] It should be noted that in the embodiment of the present application, there is no restriction on the execution subject of the vehicle scene arrangement recommendation method, and the execution subject of steps 101 to 105 can be the first vehicle or the cloud server. For the solution where the execution subject is the cloud server, the first vehicle can send the data used to express the user's current needs and the historical point data to the cloud server, and the cloud server sends the obtained first arrangement result and the second arrangement result to the first vehicle, thereby achieving the purpose of recommending vehicle scene arrangement.

[0094] The following examples describe possible implementation schemes of the vehicle scenario arrangement recommendation method described in one or more of the above embodiments.

[0095] In an embodiment of the present application, an end-to-end method of converting speech to word vectors is used, and the user's arranged speech (i.e., an example of the user's current needs) is first pre-processed to perform data cleaning and noise removal operations, and then the speech features are extracted, and the speech features are input into the speech-word vector generation model, directly converted into word embedding vectors, and finally the model reasoning is performed based on the word embedding vector to generate a speech arrangement result (i.e., an example of the first arrangement result) for execution by the vehicle end. This method omits multiple steps of traditional speech-to-text conversion, and goes directly from acoustic features to word embedding vectors. The information error in the process of converting sound to text and then to word embedding vectors is reduced, and misrecognition in the process of speech-to-text conversion is effectively avoided. The end-to-end word vector generation model can learn the direct mapping relationship between acoustic features and word embeddings, and optimize link performance.

[0096] It can be understood that in the vehicle scene arrangement recommendation method provided in the embodiment of the present application, two models are involved, namely the speech-word vector model and the business model. The speech-word vector model directly generates a word embedding vector that the business model can directly understand from the user's speech, and then passes the word embedding vector to the business model for processing to generate business logic code. Among them, the speech-word vector model belongs to the small model and is the target that needs to be optimized.

[0097] Figure 2 A schematic diagram showing a process comparison between the vehicle scene arrangement recommendation method provided in the embodiment of the present application and the general method; Figure 2 As shown, process 201 is a process of a general method, in which the user's voice is input through a voice input device 202 to obtain original voice data, the corresponding voice features are obtained through a voice feature preprocessing module 203, and the voice features are mapped into text data through a voice feature conversion module 204. The text data is passed through a large language model word segmenter 205 to obtain a word embedding vector that can be used by a large business model 206 (that is, a first recommendation model obtained by pre-training), and finally the large business model 206 recognizes the word embedding vector to generate a business logic code (that is, an example of the first arrangement result) and sends it to the vehicle end for execution.

[0098] like Figure 2As shown, process 301 is a modified version proposed in this application, in which the following modules are included: speech feature preprocessing module 203, speech-word vector model 302, business model 206 and vehicle-side execution module 207; wherein, speech feature preprocessing module 203 is used to remove unnecessary factors such as speech noise and feature extraction. Speech-word vector model 302 is used to convert speech features into word embedding vectors that can be understood by business model 206. The word embedding vector is input into business model 206 to generate business logic code. Vehicle-side execution module 207 is used to execute the business logic code generated by business model 206.

[0099] After the user's voice passes through the voice input device 202, the original voice data is obtained, and the corresponding voice features are obtained through the voice feature preprocessing module 203. Then, the voice features are directly input into the voice-word vector model 302 to generate a word embedding vector that can be used by the business model 206. The business model 206 recognizes the word embedding vector to generate business logic code and sends it to the vehicle end for execution.

[0100] It can be seen that compared with process 201, the modification scheme shown in process 301 shortens the length of the data processing link, and because the process of translating voice features into text data via the voice feature conversion text module is omitted, it is beneficial to reduce semantic errors in the voice conversion process and reduce the misrecognition rate, which in turn helps to optimize the business performance of the vehicle scenario orchestration service.

[0101] Figure 3 A schematic diagram of the steps for training and fine-tuning the general recommendation model provided in the embodiment of the present application; Figure 3 As shown, step 1 is data preparation. After sorting and cleaning all users' arrangement data, a small part of data that is reasonable and highly practical is selected as recommendation-specific data; then the cleaned data is generalized to reach the number required for model training as training data for the general recommendation model. Step 2 is the training process of the general recommendation model. The general recommendation model is trained in a supervised learning manner with the data selected and generalized in step 1 to obtain a general recommendation model with recommendation capabilities (i.e., the first recommendation model obtained by pre-training). Step 3 is the fine-tuning process of the general recommendation model. In order to complete this process, data embedding is required on the vehicle side to record the user's actual usage scenarios. This process requires periodic updates to user data. When data conditions permit, the general recommendation model obtained in step 2 is fine-tuned through the LORA fine-tuning method to obtain the user's personalized parameters (i.e., LORA parameters).

[0102] It can be understood that in the embodiment of the present application, the historical arrangement data of all users are processed, and the general recommendation model (i.e., the first recommendation model) is trained to guide new users to use the vehicle scene arrangement function. According to the usual habits and actual usage of each car owner, data statistics are performed through data burial points, and the general recommendation model is periodically fine-tuned by LORA, so that a recommendation model with personalized characteristics can be trained. In this way, the general recommendation model is used to lower the usage threshold for new users, improve the popularity and ease of use of the service arrangement function, and the use of the fine-tuned personalized recommendation model can better meet the private customization needs of car owners and improve the intelligence level of the vehicle and user satisfaction.

[0103] In the embodiments of this application, data preparation, model selection,

[0104] 1. Data preparation:

[0105] In this solution, business-related voice data is used to construct a data set. Fine-tune the word embedding generation model using business data to make it more sensitive and accurate to business voice. During the data preparation process, high-quality voice data related to the orchestration business is obtained through voice input devices, and the voice is saved through the data annotation method of voice-voice feature-text-word vector, and a certain proportion of data is selected from the data as a test set and a verification set. Among them, the data annotation method of "voice-voice feature-text-word vector" refers to the use of "voice data, voice feature data extracted after preprocessing of voice data, text data corresponding to voice data, and corresponding word embedding vector data" as a whole as the basic unit of annotated data.

[0106] 2. Model selection:

[0107] Considering the training cost and model effect, you can choose a pre-trained model of speech steering as the basic model (such as WavToVec, etc.). This type of model is pre-trained with a large amount of speech data and has the basic ability to generate speech representation vectors.

[0108] 3. Fine-tuning and testing of speech-word vector model:

[0109] In order to make the output of the above-mentioned speech-word vector model compatible with the large business model, an adaptation layer can be added after the output layer of the pre-trained speech-word vector model before fine-tuning. The input of the adaptation layer is the output of the original speech-word vector model, and the output of the adaptation layer is the same type of word vector as the input of the large business model. During the fine-tuning process, a supervised learning method is used to fine-tune the speech-word vector model. The speech feature part of the sorted business data set is input into the input layer of the fine-tuned speech-word vector model. After reasoning, the error is calculated between the result and the label, and the initial weight is modified by back propagation. After a certain number of training iterations, the effect of the speech-word vector model is evaluated and tested using the test set and the validation set, and the training strategy is adjusted according to the effect until the effect meets the business needs.

[0110] 4. Model usage:

[0111] The trained speech-word vector model and the large business model are run in series. The user's voice is transmitted to the speech feature preprocessing module through the vehicle-mounted radio equipment. The speech features output by the speech feature preprocessing module are then input into the trained speech-word vector model to obtain the word embedding vector. The word embedding vector is input into the large business model to obtain the business logic code, which is then passed to the vehicle-side execution module for business execution.

[0112] For the general recommendation model, when a user triggers a service recommended by vehicle scenario orchestration, if the vehicle-side user is a new user (i.e., a user with no embedded point data and no LORA parameter records is detected), the general recommendation model can be used in combination with the recommendation-specific data to recommend selected recommended orchestration services to the user, helping the user to quickly use the orchestration service.

[0113] For the fine-tuning model, when a user triggers a vehicle scene orchestration recommendation service, if it is detected that there is recent buried data (i.e. historical buried data) and there is a LORA parameter record, the user's recent buried data will be transmitted to the back-end program. After the back-end program pre-processes the recent buried data, it loads the general recommendation model and the LORA parameters corresponding to the user, and uses the user's recent buried data as the basic data to recommend related vehicle scene orchestration services to the user, and transmits it back to the vehicle for the user to choose.

[0114] Figure 4 A schematic diagram of a specific usage flow of the general recommendation model provided in the embodiment of the present application; Figure 4As shown in the figure, when a user triggers the vehicle scene arrangement recommendation service, the background service will query the recent tracking data of the vehicle end. If there is no historical tracking data (indicating that it is a new user), multiple recommendation results are obtained through recommendation-specific data and general recommendation models to recommend the vehicle scene arrangement service to the user, and data tracking is performed according to the recommendation result selected by the user. If the user already has historical tracking data, it means that the user has been using the vehicle for a period of time. At this time, it will be queried whether the user's LORA parameters exist. If the user does not have LORA parameters, multiple recommendation results will be obtained using recommendation-specific data and general recommendation models to recommend the vehicle scene arrangement service to the user, and data tracking is performed according to the recommendation result selected by the user. In addition, LORA is fine-tuned according to the user's historical tracking data to obtain LORA parameters; if the user's LORA parameters are queried, the LORA parameters are loaded into the general recommendation model, and the orchestration service recommendation is performed through the recent tracking data, and data tracking is performed according to the user's selection and feedback.

[0115] It should be noted that although the steps of the method in the present application are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.; or, steps in different embodiments may be combined into a new technical solution.

[0116] Based on the foregoing embodiments, an embodiment of the present application provides a vehicle scene arrangement recommendation device.

[0117] Figure 5 A schematic diagram of the structure of the vehicle scene arrangement recommendation device provided in an embodiment of the present application; Figure 5 As shown, the vehicle scene arrangement recommendation device 500 includes:

[0118] A first acquisition module 501 is configured to acquire a current demand of a user of a first vehicle;

[0119] A second acquisition module 502 is configured to acquire first historical data, wherein the first historical data includes a user's historical demand and a corresponding first control domain chain;

[0120] A first determination module 503 is configured to determine a first arrangement result for vehicle scenario arrangement by using a pre-trained first recommendation model and the current needs of the user; the first arrangement result includes a second control domain chain;

[0121] A second determination module 504 is configured to select a first control domain chain matching the current demand of the user from the first historical data as a second arrangement result for vehicle scenario arrangement;

[0122] The recommendation module 505 is configured to recommend the first arrangement result and the second arrangement result.

[0123] In some embodiments, obtaining the current needs of the user of the first vehicle includes: obtaining voice data for the first vehicle, the voice data being used to express the current needs of the user; using a pre-trained first recommendation model and the current needs of the user to determine a first arrangement result for the vehicle scene arrangement, including: performing voice feature preprocessing on the voice data to obtain voice features; inputting the voice features into a voice-word vector model to obtain a word embedding vector; and determining the first arrangement result using the pre-trained first recommendation model and the word embedding vector.

[0124] In some embodiments, the determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector without querying the LORA parameters of the first recommendation model.

[0125] In some embodiments, the determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: when the LORA parameters of the first recommendation model are queried, determining the first arrangement result by using the pre-trained first recommendation model, the LORA parameters and the word embedding vector.

[0126] In some embodiments, the obtaining of the first historical data includes: when no historical tracking data of the first vehicle is found, using pre-selected recommended special data as the first historical data; the recommended special data includes user historical demands of multiple different vehicles and the corresponding first control domain chain.

[0127] In some embodiments, the obtaining of the first historical data includes: when historical burial point data of the first vehicle is queried, using the historical burial point data as the first historical data.

[0128] In some embodiments, the first determination module 503 is further configured to: when the LORA parameters of the first recommendation model are not found, perform LORA fine-tuning on the first recommendation model according to the historical burial point data to obtain the LORA parameters of the first recommendation model.

[0129] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the device embodiment of the present application, please refer to the description of the method embodiment of the present application for understanding.

[0130] It should be noted that the division of modules in the embodiments of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional unit in each embodiment of the present application may be integrated into a processing unit, or may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of a software functional unit. It may also be implemented in the form of a combination of software and hardware.

[0131] It should be noted that in the embodiment of the present application, if the above method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present application can be essentially or partly embodied in the form of a software product that contributes to the relevant technology. The computer software product is stored in a storage medium, including several instructions to enable an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a magnetic disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific combination of hardware and software.

[0132] An embodiment of the present application provides an electronic device.

[0133] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application; Figure 6 As shown, the electronic device 600 includes a memory 601 and a processor 602, wherein the memory 601 stores a computer program that can be run on the processor 602, and the processor 602 implements the steps in the method provided in the above embodiment when executing the program.

[0134] It should be noted that the memory 601 is configured to store instructions and applications executable by the processor 602, and can also cache data to be processed or processed by the processor 602 and various modules in the electronic device 600 (for example, image data, audio data, voice communication data, and video communication data), which can be implemented through flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0135] In the embodiments of the present application, there is no limitation on the type of electronic device, and the electronic device may be a variety of devices with computing capabilities. For example, the electronic device may be a vehicle-mounted terminal, a smart phone, a laptop, a tablet computer, a cloud server, etc.

[0136] An embodiment of the present application further provides a car, comprising a memory and a processor, wherein the memory stores a computer program executable on the processor, and when the processor executes the program, the method described in the embodiment of the present application is implemented.

[0137] An embodiment of the present application also provides a computer-readable storage medium for storing a computer program.

[0138] Optionally, the computer-readable storage medium can be applied to the electronic device in the embodiments of the present application, and the computer program enables the processor or the electronic device or the automobile to execute the various methods of the embodiments of the present application, which will not be described in detail here for the sake of brevity.

[0139] An embodiment of the present application also provides a computer program product, including computer program instructions.

[0140] Optionally, the computer program product may be applied to the electronic device in the embodiments of the present application, and the computer program instructions enable the processor or the electronic device or the automobile to execute the various methods of the embodiments of the present application, which will not be described in detail here for the sake of brevity.

[0141] The embodiment of the present application also provides a computer program.

[0142] Optionally, the computer program may be applied to the electronic device in the embodiments of the present application. When the computer program runs on a processor, an electronic device, or a car, the processor or the electronic device executes the various methods of the embodiments of the present application. For the sake of brevity, they are not described here in detail.

[0143] It should be noted here that the description of the above electronic device, automobile, storage medium, computer program product and computer program embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the electronic device, automobile, storage medium, computer program product and computer program embodiment of this application, please refer to the description of the method embodiment of this application for understanding.

[0144] It should be understood that "one embodiment" or "an embodiment" or "some embodiments" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in one embodiment" or "in some embodiments" appearing throughout the specification may not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of the above-mentioned processes does not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application. The above-mentioned sequence numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments. The above description of each embodiment tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. For the sake of brevity, this article will not repeat them.

[0145] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there may be three relationships. For example, object A and / or object B can represent three situations: object A exists alone, object A and object B exist at the same time, and object B exists alone.

[0146] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0147] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as: multiple modules or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or modules can be electrical, mechanical or other forms.

[0148] The modules described above as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules; they may be located in one place or distributed on multiple network units; some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0149] In addition, all functional modules in the embodiments of the present application may be integrated into one processing unit, or each module may be a separate unit, or two or more modules may be integrated into one unit; the above-mentioned integrated modules may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0150] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0151] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling an electronic device to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0152] The methods disclosed in several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0153] The features disclosed in several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.

[0154] The features disclosed in several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0155] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A vehicle scene arrangement recommendation method, characterized in that: The vehicle scene arrangement recommendation method comprises: Obtaining current demand of a user of the first vehicle; Acquire first historical data, where the first historical data includes user historical demands and corresponding first control domain chains; Determine a first arrangement result for vehicle scenario arrangement by using a pre-trained first recommendation model and the current needs of the user; the first arrangement result includes a second control domain chain; Selecting a first control domain chain matching the current demand of the user from the first historical data as a second arrangement result for vehicle scenario arrangement; The first arrangement result and the second arrangement result are recommended.

2. The vehicle scene arrangement recommendation method according to claim 1, characterized in that: The obtaining of the current demand of the user of the first vehicle includes: obtaining voice data for the first vehicle, wherein the voice data is used to express the current demand of the user; The determining a first arrangement result for vehicle scenario arrangement by using the pre-trained first recommendation model and the current demand of the user includes: Performing speech feature preprocessing on the speech data to obtain speech features; Inputting the speech feature into a speech-word vector model to obtain a word embedding vector; The first arrangement result is determined by using a pre-trained first recommendation model and the word embedding vector.

3. The vehicle scene arrangement recommendation method according to claim 2, characterized in that: The determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: When the LORA parameters of the first recommendation model are not queried, the first arrangement result is determined by using the pre-trained first recommendation model and the word embedding vector.

4. The vehicle scene arrangement recommendation method according to claim 2, characterized in that: The determining the first arrangement result by using the pre-trained first recommendation model and the word embedding vector includes: When the LORA parameters of the first recommendation model are queried, the first arrangement result is determined using the pre-trained first recommendation model, the LORA parameters and the word embedding vector.

5. The vehicle scene arrangement recommendation method according to any one of claims 1 to 4, characterized in that: The obtaining of the first historical data includes: In the case that the historical tracking data of the first vehicle is not found, the pre-selected recommended dedicated data is used as the first historical data; the recommended dedicated data includes user historical demands of multiple different vehicles and the corresponding first control domain chain.

6. The vehicle scene arrangement recommendation method according to any one of claims 1 to 4, characterized in that: The obtaining of the first historical data includes: When the historical location data of the first vehicle is found, the historical location data is used as the first historical data.

7. The vehicle scene arrangement recommendation method according to claim 6, characterized in that: The method further comprises: If the LORA parameters of the first recommendation model are not found, the first recommendation model is fine-tuned by LORA according to the historical burial data to obtain the LORA parameters of the first recommendation model.

8. A vehicle scene arrangement recommendation device, characterized in that: The vehicle scene arrangement recommendation device comprises: A first acquisition module, configured to acquire a current demand of a user of a first vehicle; A second acquisition module is configured to acquire first historical data, wherein the first historical data includes a user's historical demand and a corresponding first control domain chain; A first determination module is configured to determine a first arrangement result for vehicle scenario arrangement by using a pre-trained first recommendation model and the current needs of the user; the first arrangement result includes a second control domain chain; A second determination module is configured to select a first control domain chain matching the current demand of the user from the first historical data as a second arrangement result for vehicle scenario arrangement; A recommendation module is configured to recommend the first arrangement result and the second arrangement result.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the vehicle scene arrangement recommendation method according to any one of claims 1 to 7 is implemented.

10. An automobile, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the vehicle scene arrangement recommendation method according to any one of claims 1 to 7 is implemented.

Images