A Deep Learning-Based Intelligent Recommendation System
By implementing a deep learning-based intelligent recommendation system on the vehicle's infotainment system, the problems of slow response speed and insufficient information security in cloud-based intelligent recommendation systems have been solved, realizing intelligent recommendation products on the vehicle's infotainment system and improving the accuracy and security of recommendations.
Patent Information
- Application Number
- CN202210755143.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-06-29
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2042-06-29
AI Technical Summary
Existing cloud-based intelligent recommendation systems in the field of connected vehicles suffer from problems such as slow response speed and insufficient information security, especially when data from the vehicle's terminal cannot be uploaded, resulting in inaccurate recommendations.
A deep learning-based intelligent recommendation system is adopted. By implementing AI algorithm preprocessing and data collection on the vehicle's infotainment system, the accuracy and data security of cloud-based intelligent recommendations are improved, enabling intelligent recommendation products on the vehicle's infotainment system. The edge computing capabilities of the vehicle's infotainment system are used for data processing and training.
It enables intelligent recommendations that respond quickly without a network connection, and ensures the accuracy of data processing while taking information security into consideration, thereby improving the accuracy and security of recommendations.
Smart Images

Figure CN117349537B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent recommendation technology, and more specifically to an intelligent recommendation system based on deep learning. Background Technology
[0002] Today, artificial intelligence is widely used in the field of connected vehicles, such as cloud-based deep learning-based intelligent recommendations in applications like online music, online radio, navigation, and vertical search. Through intelligent recommendations, online music can push songs that users like; navigation can provide users with nearby restaurants and attractions; it can remind users to pay attention to safety when they are fatigued while driving; and through intelligent recommendations, various applications can combine user profiles to provide services that are more in line with their behavioral habits.
[0003] Cloud-based intelligent recommendation systems generally suffer from several problems:
[0004] 1. Due to its reliance on the network, the response speed is not timely enough. To solve this problem, the usual solution is to deploy a rule-based intelligent recommendation engine on the vehicle's infotainment system (HU). However, this method can only achieve intelligent recommendations in certain simple scenarios.
[0005] 2. Due to information security considerations, certain data (such as GPS data) cannot be uploaded to the cloud for analysis, leading to inaccurate recommendations. To solve this problem, the usual solution is to perform information anonymization in the cloud to improve information security. However, the anonymized information has reduced accuracy, resulting in less accurate recommendations. Summary of the Invention
[0006] This invention provides a deep learning-based intelligent recommendation system that enables intelligent recommendation products on the vehicle's infotainment system. By using AI algorithms on the vehicle's infotainment system to preprocess signals, the accuracy and data security of cloud-based intelligent recommendations are improved.
[0007] The present invention discloses a deep learning-based intelligent recommendation system, comprising: an in-vehicle infotainment system and a cloud platform, wherein the in-vehicle infotainment system and the cloud platform are connected by bidirectional signals.
[0008] The in-vehicle infotainment system includes: a unified touchpoint module, an AI engine, and a unified data acquisition module.
[0009] The unified touchpoint module is responsible for the unified management of intelligent recommendation touchpoints and the priority management of in-vehicle scenarios;
[0010] The AI engine is used to submit in-vehicle scene and in-vehicle algorithm models to the cloud-based AI engine platform, including: in-vehicle scene engine and in-vehicle algorithm container.
[0011] Based on the results of the in-vehicle scene engine, the in-vehicle scene engine recommends intelligent in-vehicle scenes to the unified touchpoint module;
[0012] The vehicle-mounted algorithm container is responsible for specific algorithm services, storage of vehicle-mounted algorithm models, and inference of vehicle-mounted algorithm models;
[0013] The unified data acquisition module is responsible for uniformly collecting signals from the vehicle terminal and providing a signal input adaptation layer for the AI algorithm container. The vehicle terminal signals include: vehicle signals, system-level signals, application signals, service signals, SDS signals, and embedded signal signals.
[0014] The cloud includes a cloud AI engine platform that receives in-vehicle scene data and in-vehicle algorithm models, and combines them with cloud scene data and cloud algorithm models to train and publish cloud algorithm models.
[0015] The cloud-based AI engine platform includes: a cloud-based scene engine, a cloud-based AI algorithm container, and a big data processing module;
[0016] The cloud-based scene engine provides intelligent scene recommendations to the vehicle's infotainment system based on the results of the cloud-based AI algorithm container.
[0017] The cloud-based AI algorithm container trains a cloud-based algorithm model based on big data. After combining the cloud-based algorithm model with the vehicle-mounted algorithm model, the cloud-based AI algorithm container trains the vehicle-mounted algorithm model and then sends it to the vehicle-mounted algorithm container.
[0018] The big data processing module is responsible for big data processing.
[0019] Preferably, the unified touchpoint module manages the priority of in-vehicle scenarios according to a preset priority strategy. When the terminal intelligence recommends multiple application scenarios at the same time, the unified touchpoint module selects the scenario with higher confidence and pushes it to the user according to an arbitration strategy.
[0020] Preferably, the vehicle-mounted scene engine manages the priority of vehicle-mounted scenes according to a scene priority strategy. When the terminal intelligence recommends multiple application scenarios at the same time, the vehicle-mounted scene engine selects the scenario with higher confidence according to an arbitration strategy and pushes it to the user through the unified touchpoint module.
[0021] More preferably, the scenario priority strategy is optimized and updated through a cloud-based AI engine platform.
[0022] Preferably, the vehicle-mounted scene engine has edge intelligence capabilities, enabling data training and model inference on individual vehicle-mounted scenes.
[0023] The in-vehicle scenario refers to a personalized scenario in which a self-learning algorithm is used to mine and train machine learning algorithms on the in-vehicle device.
[0024] More preferably, the in-vehicle scenarios include: one-click Go, scenario-based vehicle control, quick operation, and one-click multimedia listening.
[0025] Preferably, the cloud AI engine platform receives the vehicle-mounted algorithm model provided by the vehicle-mounted AI engine, combines the vehicle-mounted algorithm model with the cloud algorithm model, retrains the vehicle-mounted algorithm model to obtain the cloud-based personalized scene, and sends the cloud-based personalized scene to the vehicle-mounted AI engine.
[0026] More preferably, the vehicle-mounted algorithm model is obtained by the vehicle-mounted scene engine through training the vehicle-mounted user's behavioral preferences, combined with vehicle status, real-time events, and other information, using a rule-based or content-based platform algorithm.
[0027] More preferably, the platform algorithm includes collaborative filtering, hybrid features, etc.
[0028] Preferably, the unified data acquisition module of the vehicle terminal includes an Android module. The unified data acquisition module of the Android module collects data from various services and applications in the Android system, and transmits the data from the Android module to the AI Engine algorithm engine after aggregating the data.
[0029] Vehicle signal acquisition is a complex environment. By setting up an Android module for the unified data acquisition module, the unified data acquisition module can be used with the Android system and transmit the system's embedded data to the AIEngine module in real time / non-real time.
[0030] More preferably, the unified data acquisition module of the vehicle terminal also includes a Linux terminal module. The unified data acquisition module of the Linux terminal module is used to uniformly collect data from various services and applications of the Linux system and transmit the collected data to the Android terminal module through RPC such as F-DBUS.
[0031] The unified data collection module for the Android side is used to collect data from various services and applications of the Android system, and to uniformly summarize the data from the Linux side.
[0032] The unified data acquisition module transmits the data aggregated by the Android module to the AI Engine algorithm engine.
[0033] Vehicle signal acquisition is a complex environment. By setting up Linux and Android modules for the unified data acquisition module, the unified data acquisition module can be used on dual-system vehicle platforms and can transmit the embedded data from the two systems to the AI Engine module in real time / non-real time.
[0034] More preferably, the various services and applications include: online applications, local applications, system services, application services, and data provided by suppliers.
[0035] More preferably, the vehicle-mounted data acquisition unified module also includes a customized interface for processing specific signals.
[0036] This invention discloses a deep learning-based intelligent recommendation system that enables intelligent recommendation products on the vehicle-mounted system. Based on the vehicle-mounted AI algorithm, signal preprocessing improves the accuracy and data security of cloud-based intelligent recommendations. It provides a comprehensive signal acquisition solution for the AI algorithm container; this technical feature is a module that provides big data for the AI model, can operate independently, and can provide data interfaces for other applications or services. A visualized AI algorithm model training and deployment platform can be established in the cloud, uniformly managed by the backend development and operations teams. This platform manages the collected vehicle-mounted big data and provides it to the algorithm model for inference training and iterative updates, deploying and distributing it to the vehicle-mounted intelligent model, arbitration strategies, parameter configuration files, etc.
[0037] This invention provides a deep learning-based intelligent recommendation system that can perform intelligent recommendations without relying on the network for some simple scenarios, with a fast response speed. For information security reasons, data that cannot be uploaded to the cloud can be processed directly on the vehicle's infotainment system. Since data on the vehicle's infotainment system does not require anonymization, information acquisition is more accurate, leading to more precise recommendations. Furthermore, it adapts to future data security trends by researching localized privacy computing and anonymization technologies for data. Attached Figure Description
[0038] Figure 1 This is a diagram illustrating the architecture of a deep learning-based intelligent recommendation system according to the present invention.
[0039] Figure 2 Diagram of a single-system full data acquisition scheme
[0040] Figure 3 Diagram of a dual-system full signal acquisition scheme
[0041] Figure 4 This is a flowchart of a deep learning-based intelligent recommendation system according to the present invention. Detailed Implementation
[0042] To clearly describe the present invention, it will now be described in further detail with reference to the accompanying drawings.
[0043] like Figure 1 As shown, a deep learning-based intelligent recommendation system includes: a vehicle-mounted infotainment system (HU) and a cloud-based system (Cloud). The HU and Cloud are connected by a bidirectional signal.
[0044] The in-vehicle infotainment system includes: AI Launcher (unified touchpoint module), AI Engine, and Data collector (unified data collection module).
[0045] The unified touchpoint module, AI Launcher, is responsible for the unified management of intelligent recommendations and the priority management of in-vehicle scenarios.
[0046] Unified touchpoint management: including voice touchpoints, AI assistant visual touchpoints, touchpoints that launch apps or services, HMI image touchpoints, etc.
[0047] AI Engine is used to submit vehicle-mounted scenario HU cases and vehicle-mounted algorithm models to the cloud-based AI Engine platform, including: vehicle-mounted scenario engine and vehicle-mounted algorithm container.
[0048] Based on the results of the vehicle-mounted scene engine, the vehicle-mounted scene engine recommends intelligent scenes to the unified touchpoint module. The vehicle-mounted scene engine has edge intelligence capabilities, which can perform data training and model inference on individual vehicle-mounted scenes on the vehicle-mounted side.
[0049] The vehicle-mounted algorithm container is responsible for specific algorithm services, storage of vehicle-mounted algorithm models, and inference of vehicle-mounted algorithm models;
[0050] The unified data collection module, Data collector, is responsible for uniformly collecting signals from the vehicle's infotainment system and providing a signal input adaptation layer for the AI algorithm container. The vehicle's infotainment system signals include: Car Signal, Multimode system signal, Apps application signal, Services service signal, SDS signal (Scania Driver Support System Speech Dialogue System), and Buried Point signal.
[0051] The cloud includes the cloud AI engine platform, which receives in-vehicle scene and in-vehicle algorithm model, and combines them with the cloud scene and cloud algorithm model respectively to train and publish the cloud algorithm model.
[0052] The cloud-based AI engine platform includes: a cloud-based scene engine, a cloud-based AI algorithm container, and a big data processing module (Bigdata).
[0053] The cloud-based scene engine provides intelligent scene recommendations to the vehicle's infotainment system based on the results of cloud-based AI algorithm containers.
[0054] The cloud-based AI algorithm container trains a cloud-based algorithm model based on big data. After combining the cloud-based algorithm model with the vehicle-mounted algorithm model, the cloud-based AI algorithm container trains the vehicle-mounted algorithm model and then sends it to the vehicle-mounted algorithm container.
[0055] The big data processing module is responsible for big data processing.
[0056] like Figure 2 The diagram shows a full data acquisition scheme for a deep learning-based intelligent recommendation system applicable to a single system.
[0057] The unified data acquisition module on the vehicle-mounted system also includes an Android module. The unified data acquisition module on the Android module collects data from various services and applications in the Android system (such as online applications, local applications, system services, application services, and data provided by suppliers), and then transmits the aggregated data from the Android module to the AI Engine algorithm engine.
[0058] Vehicle signal acquisition is a complex environment. By setting up an Android module for the unified data acquisition module, the unified data acquisition module can be used with the Android system and transmit the system's embedded data to the AIEngine module in real time / non-real time.
[0059] like Figure 3 The diagram shows a full data acquisition scheme for a deep learning-based intelligent recommendation system applicable to dual systems.
[0060] In addition to the Android module, the unified data acquisition module for the vehicle-mounted system also includes a Linux module. The unified data acquisition module of the Linux module is used to collect data from various services and applications of the Linux system (such as online applications, local applications, system services, application services, and data provided by suppliers), and transmit the collected data to the Android module via RPC such as F-DBUS.
[0061] The unified data collection module for the Android side is used to collect data from various services and applications of the Android system, and to uniformly summarize the data from the Linux side.
[0062] The unified data acquisition module transmits the data aggregated by the Android module to the AI Engine algorithm engine.
[0063] Vehicle signal acquisition is a complex environment. By setting up Linux and Android modules for the unified data acquisition module, the unified data acquisition module can be used on dual-system vehicle platforms and can transmit the embedded data from the two systems to the AI Engine module in real time / non-real time.
[0064] In addition to Android and Linux modules, the vehicle-mounted data acquisition unified module can also set up customized interfaces for certain special signals to process applications and data from these customized interfaces.
[0065] The in-vehicle infotainment system (HU) scenarios include: one-click Go, scenario-based vehicle control, quick operation, and one-click multimedia playback.
[0066] To further illustrate this, we will take the training of facial recognition in the cloud and on the vehicle system during scenario-based vehicle control as an example.
[0067] Initial state:
[0068] The cloud-based AI engine platform has sufficient computing power. Through existing facial big data, it trains a cloud-based algorithm model, Common model, that can identify common facial features such as male, female, elderly, and children.
[0069] The cloud-based scene engine determines that the cloud-based intelligent recommendation scene is face recognition based on the cloud-based algorithm model Common model.
[0070] The cloud-based AI engine platform stores cloud-based algorithm models, common models, and facial recognition data.
[0071] The vehicle-mounted system uses the cloud-based common model as its initial algorithm model.
[0072] Face recognition training status:
[0073] Due to the low AI computing power of in-vehicle systems, the system collects user facial data through cameras and performs transfer training using Commonmodel to train a personalized model that matches the car owner. This model can accurately identify the car owner and make personalized intelligent recommendations that match the car owner's behavior habits.
[0074] The vehicle's infotainment system uploads the trained model and facial recognition scenario to the cloud.
[0075] The cloud-based system combines the trained model and scene from the vehicle's infotainment system with the cloud-based model and scene, and then retrains the system based on big data to form a new Common model, which is then sent to the vehicle's infotainment system.
[0076] To further illustrate this, we will take One-Click Go training in the cloud and on the vehicle's infotainment system as an example.
[0077] Similar to facial recognition solutions, the cloud uses data on common actions users take when getting into the car, such as turning on the air conditioning, turning on the radio, adjusting the seat, and opening the windows, to train a common model. This model can predict the user's behavior after getting into the car under normal circumstances.
[0078] The cloud distributes the model to the vehicle's infotainment system as its initial model.
[0079] The in-vehicle infotainment system uses this initial Common model to collect data on the current driver's habits, such as setting the air conditioning temperature, turning on the radio, adjusting the seat to suit the driver, and opening the windows halfway. This data is then used for transfer training to create a personalized model that matches the current driver's preferences. The system then makes recommendations to the user and sends the trained personalized model to the cloud.
[0080] Unified touchpoint management within the unified touchpoint module includes voice touchpoints, AI assistant avatar touchpoints, app or service launch touchpoints, HMI image touchpoints, and more. For example, in a song recommendation scenario: the system interacts with the user via voice to confirm song recommendations; upon receiving a positive response, the system launches a multimedia application to play the song, and the AI assistant avatar performs interactive actions such as dancing.
[0081] The unified touchpoint module manages scene priorities by retrieving priority strategies from a pre-defined configuration file. For example, recommendations can be made according to the order of recommendation time or according to the priority level of the scene. Suppose there are two scene recommendations: one recommending songs (low priority) and the other recommending nearby gas stations (high priority); the latter can be recommended first.
[0082] The configuration file can be distributed from the cloud or be existing on the vehicle's infotainment system.
[0083] The scenario priority strategy can be to optimize and update algorithms through a cloud-based AI engine platform, or to use self-learning algorithms on the vehicle's infotainment system to mine and train machine learning algorithms for personalized scenarios.
[0084] When performing algorithm optimization and updates through the cloud AI engine platform, the cloud AI engine platform receives the vehicle-mounted algorithm model provided by the vehicle-mounted AI engine. After combining the vehicle-mounted algorithm model with the cloud algorithm model, the vehicle-mounted algorithm model is retrained to obtain the cloud-based personalized scene, and then the cloud-based personalized scene is sent to the vehicle-mounted AI engine. The vehicle-mounted algorithm model is obtained by the vehicle-mounted scene engine through training the vehicle-mounted user's behavioral preferences, combined with vehicle status, real-time events and other information, using rule-based or content-based platform algorithms (such as collaborative filtering, hybrid features, etc.).
Claims
1. A deep learning-based intelligent recommendation system, characterized in that, include: The vehicle-mounted system and the cloud are connected by a two-way signal. The in-vehicle infotainment system includes: a unified touchpoint module, an AI engine, and a unified data acquisition module. The unified touchpoint module is responsible for the unified management of intelligent recommendation touchpoints and the priority management of in-vehicle scenarios; The AI engine is used to submit in-vehicle scene and in-vehicle algorithm models to the cloud-based AI engine platform, including: in-vehicle scene engine and in-vehicle algorithm container. Based on the results of the in-vehicle scene engine, the in-vehicle scene engine recommends intelligent in-vehicle scenes to the unified touchpoint module; The vehicle-mounted algorithm container is responsible for specific algorithm services, storage of vehicle-mounted algorithm models, and inference of vehicle-mounted algorithm models; The unified data acquisition module is responsible for uniformly collecting signals from the vehicle terminal and providing a signal input adaptation layer for the AI algorithm container. The vehicle terminal signals include: vehicle signals, system-level signals, application signals, service signals, SDS signals, and embedded signal signals. The cloud includes a cloud AI engine platform that receives in-vehicle scene data and in-vehicle algorithm models, and combines them with cloud scene data and cloud algorithm models to train and publish cloud algorithm models. The cloud-based AI engine platform includes: a cloud-based scene engine, a cloud-based AI algorithm container, and a big data processing module; The cloud-based scene engine provides intelligent scene recommendations to the vehicle's infotainment system based on the results of the cloud-based AI algorithm container. The cloud-based AI algorithm container trains a cloud-based algorithm model based on big data. After combining the cloud-based algorithm model with the vehicle-mounted algorithm model, the cloud-based AI algorithm container trains the vehicle-mounted algorithm model and then sends it to the vehicle-mounted algorithm container. The big data processing module is responsible for big data processing; The cloud-based AI engine platform receives the vehicle-mounted algorithm model provided by the vehicle-mounted AI engine, combines the vehicle-mounted algorithm model with the cloud-based algorithm model, retrains the vehicle-mounted algorithm model to obtain a personalized cloud-based scene, and sends the personalized cloud-based scene to the vehicle-mounted AI engine.
2. The intelligent recommendation system based on deep learning as described in claim 1, characterized in that, The unified touchpoint module manages the priority of scenarios on the vehicle terminal according to a preset priority strategy. When the terminal intelligence recommends multiple application scenarios at the same time, the unified touchpoint module selects the scenario with higher confidence and pushes it to the user according to an arbitration strategy.
3. The intelligent recommendation system based on deep learning as described in claim 1, characterized in that, The in-vehicle scene engine prioritizes in-vehicle scenes according to a scene priority strategy. When the terminal intelligence recommends multiple application scenes at the same time, the in-vehicle scene engine selects the scene with higher confidence according to an arbitration strategy and pushes it to the user through the unified touchpoint module.
4. The intelligent recommendation system based on deep learning as described in claim 3, characterized in that, The scenario priority strategy is optimized and updated through a cloud-based AI engine platform.
5. The intelligent recommendation system based on deep learning as described in claim 1, characterized in that, The in-vehicle scenario refers to a personalized scenario in which a self-learning algorithm is used to mine and train machine learning algorithms on the in-vehicle device.
6. The intelligent recommendation system based on deep learning as described in claim 5, characterized in that, The in-vehicle infotainment scenarios include: one-click Go, scenario-based vehicle control, quick operation, and one-click multimedia listening.
7. The intelligent recommendation system based on deep learning as described in claim 1, characterized in that, The in-vehicle system algorithm model is obtained by training the in-vehicle system scene engine with the user's behavioral preferences, vehicle status, real-time events, and other information using a rule-based or content-based platform algorithm.
8. The intelligent recommendation system based on deep learning as described in claim 7, characterized in that, The platform algorithms include collaborative filtering and hybrid features.
9. The intelligent recommendation system based on deep learning as described in claim 1, characterized in that, The unified data acquisition module on the vehicle terminal includes an Android module. The unified data acquisition module of the Android module collects data from various services and applications in the Android system, and transmits the data from the Android module to the AI Engine algorithm engine.
10. The intelligent recommendation system based on deep learning as described in claim 9, characterized in that, The unified data acquisition module of the vehicle terminal also includes a Linux terminal module. The unified data acquisition module of the Linux terminal module is used to uniformly collect data from various services and applications of the Linux system and transmit the collected data to the Android terminal module through RPC such as F-DBUS. The unified data collection module for the Android side is used to collect data from various services and applications of the Android system, and to uniformly summarize the data from the Linux side. The unified data acquisition module transmits the data aggregated by the Android module to the AI Engine algorithm engine.
11. The intelligent recommendation system based on deep learning as described in claim 9, characterized in that, The data for these various services and applications includes: online applications, local applications, system services, application services, and data provided by suppliers.
Citation Information
Patent Citations
Resource processing method and device of mobile terminal, computer equipment and storage medium
CN111651263A
