Vehicle-mounted intelligent commercial promotion method and device based on AI

Abstract

According to the AI-based vehicle-mounted intelligent commercial promotion method and device provided by the embodiment of the invention, the accurate recognition of the space-time behavior mode is realized by innovatively constructing the user portrait mechanism and integrating the positioning data, the navigation track and the sensor data. And designing a scene-based push strategy, and establishing hierarchical business information indexes for intelligent matching in combination with external environment data and business resource distribution. A multi-model fusion mechanism is introduced, and multi-objective optimization of personalized push content is realized through collaborative filtering, content filtering and a comprehensive sorting model. According to the method, the defects of the traditional technology in the aspects of user portraits, pushing opportunities, content recommendation and the like are effectively overcome, and the intelligent level and the conversion effect of vehicle-mounted commercial promotion are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to an AI-based in-vehicle intelligent business promotion method and device. Background Technology

[0002] Existing in-vehicle commercial promotion methods have significant shortcomings. Traditional systems lack a systematic approach to user profiling and struggle to effectively integrate location data, navigation trajectories, and sensor data, thus affecting the accurate prediction of user interests.

[0003] Furthermore, existing technologies suffer from bottlenecks in selecting the timing of push notifications. Most systems fail to comprehensively consider external environment and scene characteristics, lacking a multi-dimensional scene recognition mechanism, resulting in less than ideal push notification performance.

[0004] Existing systems have technical shortcomings in content recommendation. They lack the ability to perform multi-model fusion analysis of commercial information, making it difficult to achieve personalized recommendations through collaborative filtering and content filtering, thus impacting user experience. Solving these problems is crucial for improving the effectiveness of commercial promotion. Summary of the Invention

[0005] To address the problems in existing technologies, this application provides an AI-based in-vehicle intelligent commercial promotion method and device, which can effectively solve the shortcomings of traditional technologies in terms of user profiling, push timing, and content recommendation, and significantly improve the intelligence level and conversion effect of in-vehicle commercial promotion.

[0006] To solve at least one of the above problems, this application provides the following technical solution:

[0007] Firstly, this application provides an AI-based in-vehicle intelligent commercial promotion method, including:

[0008] The system collects positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal. It performs sliding time window analysis on the vehicle sensor data to generate an environmental feature vector. Driving behavior features are extracted from the positioning data and navigation trajectory data to construct a driving behavior feature model. Based on the driving behavior feature model, it performs driving habit clustering analysis on the user. Combined with the environmental feature vector, it constructs a spatiotemporal behavior pattern recognition model. The output of the spatiotemporal behavior pattern recognition model is written into a user profile database. Based on the user profile database, it trains a user interest prediction model, which includes an interest decay coefficient.

[0009] Read advertising content and merchant service information from the commercial information resource database, extract feature vectors of commercial information, obtain external environment data, input the commercial information feature vectors and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push timing in different scenarios, construct a commercial resource distribution map based on the geographic information system, mark commercial outlets in layers according to road type, and generate a layered commercial information index table.

[0010] The system obtains the real-time location coordinates of the vehicle, matches the surrounding commercial information in the commercial resource distribution map, inputs the commercial information into a collaborative filtering model and a content filtering model to calculate a hybrid recommendation score, inputs the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into a comprehensive ranking model, the comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and sends the push content to the vehicle terminal display screen.

[0011] Furthermore, it also includes: reading real-time coordinate information from the GPS positioning module, path planning data from the navigation system, vehicle speed data from the vehicle CAN bus, and temperature and light data from the vehicle environment sensor through the data acquisition module in the vehicle terminal; performing time-series alignment of data from different sources based on the data acquisition timestamp; storing the aligned data in the terminal cache database; and using the data preprocessing module to normalize the cached data and remove outliers.

[0012] Feature extraction is performed on the vehicle environment sensor data using a sliding time window of length N. Statistical features of the data within the window are calculated to generate an environmental feature vector. The vehicle's position trajectory and speed change curve are extracted from the positioning data. The vehicle's path selection preference and driving time distribution are extracted from the navigation trajectory data. The environmental feature vector, along with the position trajectory, speed change curve, path selection preference, and driving time distribution, are input into a driving behavior feature extraction model to generate a user driving behavior feature vector.

[0013] Furthermore, it also includes: inputting the user's driving behavior feature vector into the K-means clustering algorithm, clustering users by calculating the Euclidean distance between the feature vectors, generating driving habit clustering labels, combining the environmental feature vector with the driving habit clustering labels to construct a sample dataset, training a spatiotemporal behavior pattern recognition model using a long short-term memory neural network, and outputting the user's behavior type label and the probability of behavior occurrence from the spatiotemporal behavior pattern recognition model;

[0014] The behavior type label, the probability of occurrence of the behavior, and the user's basic attribute information are written into the user profile database. A commercial information preference matrix is ​​constructed based on the user's historical click behavior data. A time decay function is introduced to calculate the timeliness weight of historical behaviors. The historical behavior data in the commercial information preference matrix is ​​weighted according to the timeliness weight. A neural collaborative filtering algorithm is used to train a user interest prediction model. The user interest prediction model outputs the predicted click probability values ​​of users for different types of commercial information.

[0015] Furthermore, it also includes: reading advertising content and merchant service information from a business information resource database, performing word segmentation on the advertising text content, converting the word segmentation results into text semantic vectors based on a word vector model, extracting visual features of advertising images, extracting the geographical coordinates, business hours, and service type tags of merchant services, and combining the text semantic vectors, the visual features, the geographical coordinates, the business hours, and the service type tags to construct a business information feature vector;

[0016] Weather data is obtained from a meteorological data interface, road congestion index is obtained from a traffic management platform, and holiday information is obtained from a calendar service. The weather data, road congestion index, and holiday information are used to construct an external environment feature vector. The commercial information feature vector and the external environment feature vector are input into a scene recognition model trained based on a deep neural network. The scene recognition model outputs a suitability score for pushing various types of commercial information in the current scene.

[0017] Furthermore, it also includes: calling the geographic information system interface to obtain electronic map data, extracting the road network topology, classifying and coding roads according to highways, urban arterial roads, secondary arterial roads, and branch roads, reading the geographical coordinates of commercial outlets, calculating the distance from each commercial outlet to the nearest road node, establishing a spatial association between commercial outlets and the nearest road node, and marking the location and coverage of commercial outlets in the electronic map based on the spatial association.

[0018] Commercial outlets are stored in a hierarchical manner. Commercial outlets at the same road level are constructed into an outlet set. The distance matrix between each commercial outlet within the outlet set is calculated. An adjacency graph of outlets is constructed based on the distance matrix. A unique identifier is assigned to each commercial outlet. The commercial outlet identifier, road level, location coordinates, coverage area, and adjacency relationship are written into the hierarchical commercial information index table.

[0019] Furthermore, it also includes: obtaining real-time GPS positioning coordinates from the vehicle terminal, retrieving commercial outlets within the current location radius R in the hierarchical commercial information index table, extracting the retrieved commercial outlet identification code, querying the corresponding commercial information content from the commercial information resource database based on the commercial outlet identification code, calculating the shortest path distance from the vehicle to each commercial outlet based on the outlet adjacency graph, and using the shortest path distance as a location relevance feature;

[0020] The commercial information content is input into a collaborative filtering model based on user historical behavior to calculate a collaborative recommendation score, and the commercial information content is input into a content filtering model based on product attributes to calculate a content recommendation score. The collaborative recommendation score and the content recommendation score are then weighted and fused together with the location relevance features to generate a hybrid recommendation score.

[0021] Furthermore, it also includes: constructing a comprehensive ranking and scoring function, assigning weight coefficients to the hybrid recommendation score, the scene suitability score, and the click probability prediction value respectively, solving for the optimal weight combination based on a multi-objective optimization algorithm, calculating the comprehensive score of the commercial information to be pushed using the optimal weight combination, normalizing the comprehensive score to obtain the ranking score, and sorting the commercial information to be pushed from high to low according to the ranking score;

[0022] Select the N highest-scoring business information items from the sorting results, extract the titles, descriptions, and images of the N business information items, adaptively scale the images according to the resolution of the vehicle terminal display screen, assemble the titles, descriptions, and scaled images into push content according to a preset layout template, and send the push content to the vehicle terminal display screen through a wireless communication network.

[0023] Secondly, this application provides an AI-based in-vehicle intelligent commercial promotion device, comprising:

[0024] The user profile construction module is used to collect positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal; perform sliding time window analysis on the vehicle sensor data to generate environmental feature vectors; extract driving behavior features from the positioning data and navigation trajectory data; construct a driving behavior feature model; perform driving habit clustering analysis on the user based on the driving behavior feature model; construct a spatiotemporal behavior pattern recognition model by combining the environmental feature vectors; write the output of the spatiotemporal behavior pattern recognition model into the user profile database; and train a user interest prediction model based on the user profile database, wherein the user interest prediction model includes an interest decay coefficient.

[0025] The business service determination module is used to read the advertising content and merchant service information in the business information resource database, extract the feature vector of the business information, obtain external environment data, input the business information feature vector and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push timing in different scenarios, construct a business resource distribution map based on the geographic information system, mark the business outlets in layers according to road type, and generate a layered business information index table.

[0026] The intelligent recommendation module is used to obtain the real-time location coordinates of the vehicle, match the commercial information around the current location in the commercial resource distribution map, input the commercial information into the collaborative filtering model and the content filtering model to calculate the hybrid recommendation score, and input the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into the comprehensive ranking model. The comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and the push content is sent to the vehicle terminal display screen.

[0027] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the AI-based in-vehicle intelligent business promotion method.

[0028] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the AI-based in-vehicle intelligent business promotion method.

[0029] Fifthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the AI-based in-vehicle intelligent business promotion method.

[0030] As described above, this application provides an AI-based in-vehicle intelligent commercial promotion method and device. It innovatively constructs a user profiling mechanism and integrates location data, navigation trajectory, and sensor data to achieve accurate identification of spatiotemporal behavioral patterns. A scenario-based push strategy is designed, combining external environmental data and commercial resource distribution to establish a hierarchical commercial information index for intelligent matching. A multi-model fusion mechanism is introduced, using collaborative filtering, content filtering, and a comprehensive ranking model to achieve multi-objective optimization of personalized push content. This method effectively solves the shortcomings of traditional technologies in user profiling, push timing, and content recommendation, significantly improving the intelligence level and conversion effect of in-vehicle commercial promotion. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating the AI-based in-vehicle intelligent business promotion method in the embodiments of this application;

[0033] Figure 2 This is a structural diagram of the AI-based in-vehicle intelligent commercial promotion device in the embodiments of this application;

[0034] Figure 3 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0035] Figure label:

[0036] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0038] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0039] To address the shortcomings of existing technologies, this application provides an AI-based in-vehicle intelligent commercial promotion method and device. It innovatively constructs a user profiling mechanism, integrating location data, navigation trajectories, and sensor data to achieve accurate identification of spatiotemporal behavioral patterns. A scenario-based push strategy is designed, combining external environmental data and commercial resource distribution to establish a hierarchical commercial information index for intelligent matching. A multi-model fusion mechanism is introduced, employing collaborative filtering, content filtering, and a comprehensive ranking model to achieve multi-objective optimization of personalized push content. This method effectively solves the deficiencies of traditional technologies in user profiling, push timing, and content recommendation, significantly improving the intelligence level and conversion rate of in-vehicle commercial promotion.

[0040] To effectively address the shortcomings of traditional technologies in areas such as user profiling, push notification timing, and content recommendation, and to significantly improve the intelligence level and conversion rate of in-vehicle commercial promotion, this application provides an embodiment of an AI-based in-vehicle intelligent commercial promotion method. See [link to embodiment]. Figure 1 The AI-based in-vehicle intelligent business promotion method specifically includes the following:

[0041] Step S101: Collect positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal; perform sliding time window analysis on the vehicle sensor data to generate an environmental feature vector; extract driving behavior features from the positioning data and the navigation trajectory data; construct a driving behavior feature model; perform driving habit clustering analysis on the user based on the driving behavior feature model; construct a spatiotemporal behavior pattern recognition model by combining the environmental feature vector; write the output of the spatiotemporal behavior pattern recognition model into a user profile database; train a user interest prediction model based on the user profile database; the user interest prediction model includes an interest decay coefficient.

[0042] Optionally, this embodiment addresses the problems of inaccurate user profiling, inaccurate interest prediction, and poor recommendation performance in traditional in-vehicle commercial promotion by innovatively designing a user behavior analysis scheme based on multi-source data fusion. This embodiment collects multi-dimensional data in real time through the in-vehicle terminal, including the location coordinates and timestamps of the GPS positioning module, the path planning and actual trajectory of the navigation system, the vehicle speed and acceleration of the CAN bus, and data from environmental sensors such as temperature, humidity, and light intensity. A distributed data acquisition architecture is adopted, establishing a data cache queue: Data_Queue = Buffer(GPS, Nav, CAN, Sensor), where each parameter corresponds to a real-time data stream from a different data source. A timestamp alignment mechanism ensures the temporal consistency of the data.

[0043] This embodiment deeply optimizes the environmental feature extraction mechanism. A sliding time window technique is used to process the sensor data stream, with the window length dynamically adjusted based on data characteristics: Window_Size = f(Sample_Rate, Feature_Type), where Sample_Rate is the sampling frequency and Feature_Type is the feature type. Within each time window, statistical features, including mean, variance, kurtosis, and skewness, are calculated to construct a multi-dimensional environmental feature vector. For example, analysis of illumination data can reflect the diurnal variation of driving scenarios, while the combined features of temperature and humidity reflect the impact of weather conditions on driving behavior.

[0044] This embodiment innovatively implements a driving behavior feature extraction strategy. It analyzes driving route selection preferences based on location trajectory data and identifies frequently used routes through trajectory clustering. Time-frequency domain analysis is performed on speed curves to extract driving operation features such as acceleration, deceleration, and cruising. The system combines the deviation between the navigation path and the actual trajectory to assess the driver's path adherence. Through the combined expression of multi-dimensional features, a complete driving behavior profile is constructed, providing a foundation for subsequent personalized recommendations.

[0045] This embodiment optimizes the driving habit clustering analysis mechanism. An improved K-means algorithm is used to cluster driving behavior features, and the optimal number of clusters is determined adaptively through silhouette coefficients. The system considers feature weights in both time and space dimensions, and optimizes clustering performance through distance metrics. For example, for commuting scenarios, the focus is on the stability of route selection; for leisure travel, more emphasis is placed on driving style characteristics.

[0046] This embodiment innovatively designs a spatiotemporal behavior pattern recognition scheme. A behavior recognition model is constructed based on a Long Short-Term Memory (LSTM) network, with the input layer containing environmental feature vectors and driving habit clustering labels. The network structure captures long-term dependencies through an attention mechanism, recognizing user behavior patterns under different spatiotemporal conditions. The model output includes a behavior type probability distribution and temporal feature representation, supporting the dynamic updating of user profiles.

[0047] This embodiment deeply optimizes the user profile construction mechanism. It employs multimodal feature fusion technology to integrate behavioral recognition results with basic user attributes. The system designs a hierarchical profile structure, including static features, dynamic preferences, and temporal behavioral patterns. Knowledge graph technology is used to establish relationships between features, enhancing the expressive power and interpretability of the profile.

[0048] This embodiment innovatively implements an interest prediction model. Based on a neural collaborative filtering framework, a time decay function is introduced: Decay(t) = exp(-λ×Δt), where λ is the decay coefficient and Δt is the time interval. The system trains a deep learning model using historical behavioral data to predict users' click preferences for different types of commercial information. The model specifically considers the timeliness of interests, dynamically adjusting the weights of historical behaviors through a decay mechanism.

[0049] This embodiment achieves accurate modeling of user behavior through multi-source data analysis technology. It demonstrates particularly strong feature extraction and pattern recognition capabilities when handling complex driving scenarios. Through the application of deep learning models, the system can accurately capture user behavioral characteristics and interests, providing a reliable basis for business promotion decisions.

[0050] This embodiment's innovative design not only solves the problems of user profiling accuracy and interest prediction accuracy in traditional methods, but also establishes a continuously optimizeable user behavior analysis framework. Through continuous optimization of the analysis model and improvement of prediction strategies, the system can continuously enhance its understanding of user behavior, providing strong support for intelligent business promotion. This AI-based analysis mechanism ensures that the system maintains efficient modeling capabilities and reliable prediction results when facing dynamically changing user behavior. In particular, the application of multi-dimensional data fusion and deep learning models significantly improves the accuracy of user profiling and the effectiveness of personalized recommendations.

[0051] Step S102: Read the advertising content and merchant service information in the commercial information resource database, extract the feature vector of the commercial information, obtain external environment data, input the commercial information feature vector and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push time in different scenarios, construct a commercial resource distribution map based on the geographic information system, mark the commercial outlets in layers according to road type, and generate a layered commercial information index table.

[0052] Optionally, this embodiment addresses the problems of incomplete content feature extraction, inaccurate scene recognition, and unreasonable resource distribution management in traditional in-vehicle commercial promotion by innovatively designing a commercial information processing solution based on multimodal analysis. This embodiment first establishes a data interface for a commercial information resource database, employing a distributed storage architecture to manage advertising content and merchant information. It then uses a natural language processing engine to perform deep analysis of the text content, extracting semantic features using an improved Word2Vec model: Text_Vector = Embedding(Content, Context), where Content is the text content and Context is the context information, ensuring that the feature vector accurately expresses the semantic connotation of the commercial information.

[0053] This embodiment deeply optimizes the commercial information feature extraction mechanism. For advertising image content, a deep convolutional neural network is used to extract visual features, including color distribution, texture features, and object detection results. The system combines computer vision technology to analyze the quality and attractiveness of the image, and enhances feature expression capabilities through multi-layer feature fusion. For merchant service information, a complete attribute system is established, including dimensions such as geographical location, business hours, service type, and user reviews. Knowledge graph technology is used to construct the relationships between attributes, forming a structured feature representation.

[0054] This embodiment innovatively implements an environmental data processing strategy. A multi-source data acquisition network is constructed to acquire real-time external environmental data such as weather conditions, traffic status, and holiday information. The system employs a standardized data processing workflow to convert environmental data from different sources into a unified format. Particularly for dynamically changing data, such as traffic congestion indices, time-series analysis is used to predict their changing trends, providing timely support for scene recognition. For example, in rainy weather scenarios, the system pays special attention to commercial information related to restaurants and leisure venues; during holidays, it focuses more on pushing tourism and shopping information.

[0055] This embodiment optimizes the scene recognition model design. A scene recognition framework is constructed based on a deep neural network, with the model input layer containing commercial information feature vectors and environmental feature vectors. A multi-head attention mechanism is used to capture the correlation between different features and adaptively adjust the feature weights. The system designs a scene scoring function: Score = f(Content, Environment, Time), where each parameter represents content relevance, environment adaptability, and time suitability, respectively. The model outputs a push suitability score for different scenarios, providing a decision-making basis for selecting the optimal push timing.

[0056] This embodiment deeply optimizes the geographic information processing mechanism. It calls the geographic information system interface to obtain high-precision electronic map data and extracts the road network topology. The system adopts a multi-level road classification scheme, encoding and marking different types of roads such as highways, arterial roads, and secondary arterial roads. Spatial analysis algorithms are used to calculate the association between commercial outlets and road nodes, constructing a location index system. Special consideration is given to the density and coverage of commercial areas, optimizing resource distribution representation through spatial clustering.

[0057] This embodiment innovatively implements a commercial resource indexing strategy. A hierarchical index table is constructed based on a road hierarchy, and a tree structure is used to manage commercial outlet data. The system assigns a unique identifier to each commercial outlet and records its spatial location, coverage area, associated roads, and other attribute information. Spatial proximity analysis establishes relationships between outlets, supporting rapid spatial queries and route planning. For densely populated commercial areas, grid-based management improves retrieval efficiency.

[0058] This embodiment achieves precise processing of commercial information through multimodal analysis technology. It demonstrates particularly strong feature extraction capabilities and scene recognition performance when handling complex scenarios. Through the application of deep learning models, the system can accurately understand the content characteristics and scene adaptability of commercial information, providing reliable decision support for precise delivery.

[0059] This embodiment's innovative design not only solves the feature extraction and scene recognition problems of traditional methods but also establishes a continuously optimizeable business information processing framework. Through continuous optimization of the analysis model and improvement of the indexing strategy, the system can continuously enhance its management capabilities for business resources, providing strong support for intelligent promotion. This AI-based processing mechanism ensures that the system maintains efficient analytical capabilities and reliable recognition results when facing diverse business scenarios. In particular, the application of multi-dimensional data fusion and deep learning models significantly improves the accuracy of business information processing and the ability to grasp the timing of push notifications.

[0060] Step S103: Obtain the real-time location coordinates of the vehicle, match the commercial information around the current location in the commercial resource distribution map, input the commercial information into the collaborative filtering model and the content filtering model to calculate the hybrid recommendation score, input the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into the comprehensive ranking model, the comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and send the push content to the vehicle terminal display screen.

[0061] Optionally, this embodiment addresses the problems of mismatched recommended content, inappropriate push timing, and simplistic sorting strategies in traditional in-vehicle commercial promotion by innovatively designing an intelligent push solution based on multi-model fusion. This embodiment first obtains vehicle location information in real time through the vehicle terminal's GPS module, employing a high-precision positioning algorithm to ensure location accuracy. A spatial index is constructed based on a commercial resource distribution map: Index = KDTree(POI, Distance), where POI is the coordinates of the commercial outlet and Distance is the search radius. Spatial retrieval quickly locates surrounding commercial resources.

[0062] This embodiment deeply optimizes the recommendation algorithm fusion mechanism. Commercial information is simultaneously input into both collaborative filtering and content filtering models, and a weighted fusion is used to generate a hybrid recommendation score: Score = α × CF + (1-α) × CB, where CF is the collaborative filtering score, CB is the content filtering score, and α is the dynamically adjusted fusion weight. The collaborative filtering model, based on users' historical behavior data, mines common interest patterns among user groups; the content filtering model focuses on the attribute characteristics of commercial information to analyze the matching degree between users and products. The system continuously optimizes the fusion weights through online learning methods to adapt to recommendation needs in different scenarios.

[0063] This embodiment innovatively implements a scene-aware push strategy. It analyzes current scene characteristics based on a deep learning model, including multi-dimensional information such as time, location, weather, and traffic. The model identifies key factors affecting push effectiveness through an attention mechanism and outputs appropriate push times for different types of commercial information. For example, on a user's daily commute, it pushes nearby restaurant information when the vehicle is stopped or waiting at a red light; during weekend leisure driving, it pushes relevant entertainment venue information based on weather conditions.

[0064] This embodiment optimizes the user interest prediction mechanism. A neural network model is used to analyze users' historical interaction data, and feature engineering is employed to extract users' interest and preference features. The model considers the time decay effect of interests, giving higher weight to recent behaviors. A click-through rate prediction model is used to assess users' potential interest in different commercial information, providing an important basis for subsequent content ranking. The system pays special attention to the dynamic changes in user interests, updating the prediction model through a real-time feedback mechanism.

[0065] This embodiment innovatively designs a multi-objective ranking scheme. A comprehensive scoring function is constructed: Rank = w1Mix + w2Time + w3×CTR, where Mix is ​​the mixed recommendation score, Time is the scene suitability score, CTR is the predicted click probability, and w1, w2, and w3 are the corresponding weights. The system employs a multi-objective optimization algorithm to solve for the optimal weight combination, balancing the three objectives of recommendation relevance, timing appropriateness, and user interest. A genetic algorithm iteratively optimizes the weight parameters to ensure the global optimality of the ranking result.

[0066] This embodiment deeply optimizes the content generation mechanism. The final push content is filtered based on the sorting results, and intelligent layout is implemented considering the characteristics of in-vehicle displays. The system designs adaptive layout templates, adjusting the display position and size according to content type and importance. For image materials, image processing algorithms are used for compression and optimization, ensuring display quality while reducing transmission load. The system also considers driving safety factors, dynamically adjusting the complexity of information display at different vehicle speeds.

[0067] This embodiment innovatively implements a push notification monitoring strategy. A complete push notification effectiveness evaluation system is established to monitor user interaction behavior in real time. A / B testing is used to verify the effectiveness of different push notification strategies, continuously optimizing the recommendation algorithm and ranking model. Through a feedback loop mechanism, the system uses real-time user responses as a crucial basis for model optimization, continuously improving the accuracy and timeliness of push notifications.

[0068] This embodiment achieves intelligent delivery of commercial information through multi-model fusion technology. It demonstrates particularly strong scene understanding and decision-making capabilities when handling complex driving scenarios. Through the application of deep learning models, the system can accurately determine the timing of push notifications and rationally prioritize content, providing users with personalized commercial information services.

[0069] This embodiment's innovative design not only solves the problems of push notification effectiveness and user experience in traditional methods, but also establishes a continuously optimizeable push notification decision-making framework. Through continuous optimization of the algorithm model and improvement of decision-making strategies, the system can continuously improve its responsiveness to user needs, providing strong support for intelligent commercial promotion. This AI-based push mechanism ensures that the system maintains efficient decision-making capabilities and reliable push notification effects when facing dynamically changing scenarios. In particular, through multi-dimensional feature fusion and multi-objective optimization strategies, it significantly improves the conversion rate and user satisfaction of commercial promotion.

[0070] As described above, the AI-based in-vehicle intelligent commercial promotion method provided in this application can achieve accurate identification of spatiotemporal behavioral patterns by innovatively constructing a user profile mechanism and integrating location data, navigation trajectory, and sensor data. It designs a scenario-based push strategy, combining external environmental data and commercial resource distribution to establish a hierarchical commercial information index for intelligent matching. A multi-model fusion mechanism is introduced, using collaborative filtering, content filtering, and a comprehensive ranking model to achieve multi-objective optimization of personalized push content. This method effectively solves the shortcomings of traditional technologies in user profiling, push timing, and content recommendation, significantly improving the intelligence level and conversion effect of in-vehicle commercial promotion.

[0071] In one embodiment of the AI-based in-vehicle intelligent commercial promotion method of this application, it may further include the following:

[0072] Step S201: Read the real-time coordinate information of the GPS positioning module, the path planning data of the navigation system, the vehicle speed data of the vehicle CAN bus, and the temperature and light data of the vehicle environment sensor through the data acquisition module in the vehicle terminal. Based on the data acquisition timestamp, perform time-series alignment on the data from different sources, store the aligned data in the terminal cache database, and use the data preprocessing module to normalize the cache data and remove outliers.

[0073] Step S202: Apply a sliding time window of length N to the vehicle environment sensor data for feature extraction, calculate the statistical features of the data within the window to generate an environmental feature vector, extract the vehicle's position trajectory and speed change curve from the positioning data, extract the vehicle's path selection preference and driving time distribution from the navigation trajectory data, and input the environmental feature vector, the position trajectory, the speed change curve, the path selection preference, and the driving time distribution into the driving behavior feature extraction model to generate a user driving behavior feature vector.

[0074] Optionally, this embodiment addresses the problems of asynchronous multi-source data, unstable data quality, and incomplete feature extraction inherent in traditional vehicle-mounted data acquisition by innovatively designing a data processing scheme based on multi-dimensional perception. This embodiment first constructs a multi-source data acquisition architecture for the vehicle terminal, connecting to data sources such as GPS positioning modules, navigation systems, CAN buses, and environmental sensors via a high-speed data bus. A distributed data acquisition framework is adopted, establishing a synchronization mechanism for data with different sampling frequencies: Sync_Time = Base_Time + Offset, where Base_Time is the system base time and Offset is the time deviation of each data source, ensuring the temporal consistency of the data.

[0075] This embodiment deeply optimizes the data preprocessing mechanism. Multi-source data is aligned based on timestamp information, and a sliding window technique is used to achieve a smooth data transition. The system designs an adaptive anomaly detection algorithm that identifies outliers through statistical feature analysis: Anomaly_Score = |x - μ| / σ, where x is the observed value, μ is the mean, and σ is the standard deviation. Differentiated normalization strategies are used for different types of data to ensure the consistency of data distribution. For example, maximum-minimum normalization is used for speed data, while Z-score normalization is used for temperature data.

[0076] This embodiment innovatively implements an environmental data processing strategy. A sliding time window analysis is applied to environmental sensor data, with the window length N dynamically adjusted based on data characteristics. Within each window, multidimensional statistical features are calculated, including mean, variance, kurtosis, and skewness, to construct an environmental feature vector. The system pays particular attention to the changing trends of environmental parameters, capturing the dynamic characteristics of environmental conditions through time-series analysis. For example, temperature changes can reflect in-vehicle comfort, while changes in illumination indicate driving visibility conditions.

[0077] This embodiment optimizes the location trajectory analysis mechanism. Continuous position coordinate sequences are extracted from GPS positioning data, and positioning noise is eliminated through a trajectory smoothing algorithm. The system calculates the geometric features of the trajectory, including steering angle, travel distance, and dwell time. Combined with path planning data from the navigation system, the system analyzes the driver's route selection preferences, identifying frequently used routes and preferred road segments. For speed change curves, driving state features such as acceleration, deceleration, and cruising are extracted through time-frequency domain analysis.

[0078] This embodiment deeply optimizes the behavioral feature extraction mechanism. A multi-layered feature extraction model is constructed, integrating environmental feature vectors, location trajectory features, speed features, and path preference features. The system employs a deep learning network to design the feature extractor, extracting high-order feature representations through a multilayer perceptron and attention mechanism. Particular attention is paid to the correlation between different features, capturing complex behavioral patterns through a feature interaction layer. For example, by analyzing the relationship between speed changes and path selection, the driver's style preferences can be identified.

[0079] This embodiment innovatively implements a data storage strategy. A tiered storage architecture is used to manage the collected data, with differentiated storage solutions designed for different data types. The system establishes an efficient data indexing mechanism, supporting fast time-series queries and spatial retrieval. For important feature data, data compression and redundant backups ensure data reliability. Especially for historical data accumulated over a long period, storage space is optimized through data archiving and cleanup strategies.

[0080] This embodiment achieves accurate characterization of driving behavior through multi-source data fusion technology. It demonstrates particularly strong data processing capabilities and feature extraction performance when handling complex driving scenarios. Through the application of deep learning models, the system can accurately capture multi-dimensional features of driving behavior, providing a reliable data foundation for personalized recommendations.

[0081] This embodiment's innovative design not only solves the data synchronization and quality issues of traditional methods but also establishes a continuously optimizeable data processing framework. Through continuous improvement of processing strategies and optimization of feature extraction methods, the system can continuously enhance its understanding of driving behavior, providing strong support for intelligent recommendations. This multi-dimensional perception-based processing mechanism ensures that the system maintains efficient data processing capabilities and reliable feature extraction results when facing complex and ever-changing driving scenarios. In particular, the deep fusion of multi-source data significantly improves the expressive power of driving behavior features and the accuracy of model predictions.

[0082] In one embodiment of the AI-based in-vehicle intelligent commercial promotion method of this application, it may further include the following:

[0083] Step S301: Input the user's driving behavior feature vector into the K-means clustering algorithm, and cluster the user by calculating the Euclidean distance between the feature vectors to generate driving habit clustering labels. Combine the environmental feature vector with the driving habit clustering labels to construct a sample dataset. Use a long short-term memory neural network to train a spatiotemporal behavior pattern recognition model. The spatiotemporal behavior pattern recognition model outputs the user's behavior type label and the probability of behavior occurrence.

[0084] Step S302: Write the behavior type label, the probability of the behavior, and the user's basic attribute information into the user profile database. Construct a commercial information preference matrix based on the user's historical click behavior data. Introduce a time decay function to calculate the timeliness weight of historical behaviors. Perform weighted processing on the historical behavior data in the commercial information preference matrix according to the timeliness weight. Train a user interest prediction model using a neural collaborative filtering algorithm. The user interest prediction model outputs the predicted click probability values ​​of users for different types of commercial information.

[0085] Optionally, this embodiment addresses the problems of inaccurate behavior recognition, imprecise interest prediction, and insufficient consideration of timeliness in traditional user profiling by innovatively designing a user modeling scheme based on deep learning. This embodiment first processes driving behavior feature vectors using the K-means clustering algorithm, and then calculates the similarity between users using a distance metric function: Distance = sqrt(Σ(xi-yi)). 2 ), where xi and yi represent the feature vector components of two users, respectively. The system uses the silhouette coefficient method to adaptively determine the optimal number of clusters, and ensures the stability of the clustering results through iterative optimization.

[0086] This embodiment deeply optimizes the clustering analysis mechanism. An adaptive feature weight adjustment strategy is designed to address the multidimensional characteristics of driving behavior. The system considers the importance of different features and performs weighted processing on distance calculations. For example, in commuting scenarios, route selection stability has a higher weight; while in leisure driving, speed and environmental response features are more important. Through clustering analysis, users are divided into different driving habit groups, providing a foundation for subsequent behavior recognition.

[0087] This embodiment innovatively implements a spatiotemporal behavior modeling strategy. A Long Short-Term Memory (LSTM) neural network is used to construct the behavior recognition model, with the input layer fusing environmental feature vectors and driving habit clustering labels. The network structure contains multiple layers of LSTM units, capturing long-term dependencies through a gating mechanism. A specially designed attention layer dynamically allocates attention weights based on the importance of different time steps. The model output includes multi-classification results for behavior types and their corresponding probability distributions, achieving accurate identification of user behavior.

[0088] This embodiment optimizes the user profile construction mechanism. It integrates behavior recognition results with basic user attribute information to construct a multi-dimensional user profile. The system uses a graph database to store user profiles and expresses the relationships between features through an attribute graph model. Differentiated update strategies are designed for different types of attribute information to ensure the timeliness of profile data. In particular, for dynamically changing behavioral features, a real-time update mechanism maintains the accuracy of the profile.

[0089] This embodiment deeply optimizes the interest modeling mechanism. A preference matrix is ​​constructed based on users' historical click behavior, and a time decay function is introduced: Weight = exp(-λ*Δt), where λ is the decay coefficient and Δt is the time interval. The system dynamically adjusts the influence of historical behavior through the decay function, giving more weight to recent behavior. For behavioral patterns that have not appeared for a long time, their impact on the current prediction is reduced through weight decay.

[0090] This embodiment innovatively implements an interest prediction model. It employs a neural collaborative filtering framework to map user features and commercial information features to the same latent space. The model learns feature interaction patterns through multi-layer neural networks, capturing the deep correlation between user interests and commercial information. The system pays special attention to the cold start problem, using transfer learning to handle recommendation needs from new users and new products. The model outputs predicted click probabilities for different types of commercial information, providing a basis for personalized recommendations.

[0091] This embodiment achieves accurate modeling of user interests through deep learning technology. It demonstrates particularly strong pattern recognition and prediction capabilities when handling complex behavioral sequences. By applying an LSTM network, the system can accurately capture the temporal characteristics of user behavior, providing reliable interest predictions for commercial push notifications.

[0092] This embodiment's innovative design not only solves the problems of behavior recognition and interest prediction in traditional methods, but also establishes a continuously optimizeable user modeling framework. Through continuous model optimization and improvement of prediction strategies, the system can continuously enhance its understanding of user behavior, providing strong support for intelligent push notifications. This deep learning-based modeling mechanism ensures that the system maintains efficient recognition capabilities and reliable prediction results when facing dynamically changing user behavior. In particular, the introduction of a time decay mechanism significantly improves the timeliness and accuracy of interest prediction.

[0093] In one embodiment of the AI-based in-vehicle intelligent commercial promotion method of this application, it may further include the following:

[0094] Step S401: Read the advertising content and merchant service information from the business information resource database, perform word segmentation on the advertising text content, convert the word segmentation results into text semantic vectors based on the word vector model, extract the visual features of the advertising image, extract the geographical coordinates, business hours, and service type tags of the merchant services, and combine the text semantic vector, the visual features, the geographical coordinates, the business hours, and the service type tags to construct a business information feature vector.

[0095] Step S402: Obtain weather data from the meteorological data interface, obtain road congestion index from the traffic management platform, and obtain holiday information from the calendar service. Construct an external environment feature vector using the weather data, the road congestion index, and the holiday information. Input the commercial information feature vector and the external environment feature vector into a scene recognition model trained based on a deep neural network. The scene recognition model outputs a suitability score for pushing various types of commercial information in the current scene.

[0096] Optionally, this embodiment addresses the problems of incomplete feature extraction, inaccurate scene recognition, and imprecise timing of push notifications in traditional business information processing by innovatively designing a business information processing solution based on multimodal analysis. This embodiment first constructs a data access interface for a business information resource database, employing a distributed storage architecture to manage advertising content and merchant information. It then performs deep analysis of the text content using a natural language processing engine, employing an improved BERT model for word segmentation and semantic feature extraction: Text_Vector = BERT(Content, Context), where Content is the text content and Context is the context information, ensuring that the semantic vector accurately represents the content features of the business information.

[0097] This embodiment deeply optimizes the multimodal feature extraction mechanism. For advertising image content, a convolutional neural network is used to extract visual features, including color distribution, texture features, and object detection results. The system optimizes the visual feature extractor through transfer learning, paying particular attention to key visual elements in commercial scenarios. For merchant service information, a complete attribute system is established, including spatial representation of location coordinates, periodic features of business hours, and semantic encoding of service types. Knowledge graph technology is used to construct the relationships between attributes, forming a structured feature representation: Feature = Concat(Text, Visual, Location, Time, Service), where each component represents text features, visual features, location features, time features, and service features, respectively.

[0098] This embodiment innovatively implements an environmental data processing strategy. A multi-source data acquisition network is constructed to acquire real-time external environmental data such as weather conditions, traffic status, and holiday information. The system employs a data standardization processing flow to convert environmental data from different sources into a unified format. For example, weather conditions are encoded to include factors such as temperature, humidity, and precipitation; traffic congestion indices are quantified to reflect road conditions; and holiday information undergoes semantic enhancement, taking into account the impact of holiday type and duration.

[0099] This embodiment optimizes the scene feature fusion mechanism. An attention mechanism is employed to fuse business information feature vectors with environmental feature vectors, adaptively adjusting the weights of different features. The system designs a hierarchical feature interaction structure, capturing the nonlinear relationships between features through a multilayer perceptron. Particularly for time-related features, recurrent neural networks are used to model their dynamic changes, improving the accuracy of scene understanding.

[0100] This embodiment deeply optimizes the scene recognition model design. A scene recognition framework is built based on a deep neural network, and the model employs a multi-branch structure to handle different types of input features. An attention mechanism is used to achieve adaptive feature fusion, focusing on feature combinations that significantly influence push decisions. The system designs a scene scoring function that comprehensively considers content relevance, environmental adaptability, and time suitability to generate push appropriateness scores for different types of commercial information.

[0101] This embodiment innovatively implements a model training strategy. A large-scale labeled dataset is used to train the scene recognition model, and a multi-task learning method is employed to improve the model's generalization ability. The system designs a targeted loss function to balance recognition accuracy across different scenarios. An online learning mechanism continuously optimizes model parameters to adapt to dynamic changes in scene features. For example, in rainy weather scenarios, the model improves the appropriateness of pushing information related to indoor locations; during holidays, it focuses more on the timing of pushing leisure and entertainment information.

[0102] This embodiment achieves accurate identification of business scenarios through multimodal analysis technology. It demonstrates particularly strong feature extraction capabilities and scene understanding when handling complex scenarios. Through the application of deep learning models, the system can accurately determine the timing of push notifications, providing reliable decision support for precision marketing.

[0103] This embodiment's innovative design not only solves the feature extraction and scene recognition problems of traditional methods but also establishes a continuously optimizeable scene analysis framework. Through continuous optimization of the analysis model and improvement of feature extraction strategies, the system can continuously enhance its understanding of business scenarios, providing strong support for intelligent push notifications. This AI-based analysis mechanism ensures that the system maintains efficient analysis capabilities and reliable recognition results when facing diverse business scenarios. In particular, the application of multi-dimensional data fusion and deep learning models significantly improves the accuracy and timeliness of business information push notifications.

[0104] In one embodiment of the AI-based in-vehicle intelligent commercial promotion method of this application, it may further include the following:

[0105] Step S501: Call the geographic information system interface to obtain electronic map data, extract the road network topology, classify and encode the roads according to highways, urban arterial roads, secondary arterial roads and branch roads, read the geographical coordinates of commercial outlets, calculate the distance from each commercial outlet to the nearest road node, establish spatial association between commercial outlets and the nearest road node, and mark the location and coverage of commercial outlets in the electronic map based on the spatial association.

[0106] Step S502: Store commercial outlets in a hierarchical manner, construct a set of outlets for commercial outlets at the same road level, calculate the distance matrix between each commercial outlet within the set, construct an adjacency graph based on the distance matrix, assign a unique identifier to each commercial outlet, and write the commercial outlet identifier, road level, location coordinates, coverage area, and adjacency relationship into the hierarchical commercial information index table.

[0107] Optionally, this embodiment addresses the problems of inaccurate spatial positioning, unclear hierarchical relationships, and low retrieval efficiency in traditional commercial outlet management by innovatively designing a GIS-based commercial resource spatial management solution. This embodiment first acquires high-precision electronic map data through a geographic information system interface and then extracts the road network topology using graph theory methods. The system designs a multi-level road classification scheme, using the coding rule: Road_Code = Level_ID + Area_Code + Road_ID, where Level_ID is the road level identifier, Area_Code is the area code, and Road_ID is the road sequence number, thus achieving hierarchical management of the road network.

[0108] This embodiment deeply optimizes the spatial association analysis mechanism. Based on the geographic coordinates of commercial outlets, a nearest neighbor search algorithm is used to calculate the distance from the outlet to the road node. The system determines the service area of ​​each outlet through Voronoi diagram analysis and establishes a spatial mapping relationship between outlets and road nodes: Distance = sqrt((x1-x2)). 2 +(y1-y2) 2 ), where (x1, y1) and (x2, y2) are the coordinates of the network point and road node, respectively. Spatial indexing technology is used to optimize distance calculation efficiency, ensuring rapid processing of large-scale network point data.

[0109] This embodiment innovatively implements a network coverage analysis strategy. Differentiated coverage calculation methods are designed for different types of commercial outlets. For example, for large shopping malls, their radiation range and accessibility are considered; for convenience stores, walking distance coverage is emphasized. The system generates influence contour lines for outlets through spatial analysis algorithms and assesses the actual service range in conjunction with the road network structure. Particularly in densely populated commercial areas, overlap analysis optimizes outlet layout.

[0110] This embodiment optimizes the hierarchical storage mechanism. A multi-level network point index structure is constructed based on road levels, organizing commercial network points on roads of the same level into ordered sets. The system employs R-tree indexing technology to achieve efficient spatial queries, supporting range search and nearest neighbor retrieval. For each network point set, spatial clustering methods are used to identify densely populated commercial areas, providing a basis for differentiated service strategies.

[0111] This embodiment deeply optimizes the network relationship modeling mechanism. An adjacency graph is constructed based on the distance matrix between network points, and a graph database is used to store the topological relationships between network points. The system analyzes the accessibility between network points using a shortest path algorithm, considering the impact of road grade and traffic conditions. Especially for chain retail outlets, the relationship graph analyzes the coverage characteristics of their service network, supporting regional-level allocation of commercial resources.

[0112] This embodiment innovatively implements an index table construction strategy. A globally unique identifier is assigned to each business outlet: POI_ID = Type_Code + Region_Code + Serial_Number, where Type_Code is the business type code, Region_Code is the region code, and Serial_Number is the serial number. The system organizes the core attributes and relationship information of the outlets into a structured index table, supporting multi-dimensional information retrieval. A caching mechanism optimizes the access efficiency of hot data and improves the response speed of push services.

[0113] This embodiment achieves precise management of commercial resources through spatial analysis technology. It demonstrates particularly strong spatial processing capabilities and retrieval efficiency when handling large-scale network data. Through the application of a hierarchical indexing mechanism, the system can quickly locate commercial resources in target areas, providing spatial decision support for precise targeting.

[0114] This embodiment's innovative design not only solves the spatial management and retrieval efficiency problems of traditional methods but also establishes a sustainable and optimized resource management framework. Through continuous optimization of indexing strategies and improvements in spatial analysis methods, the system can continuously enhance its management capabilities for commercial resources, providing strong support for intelligent push notifications. This GIS-based management mechanism ensures that the system maintains efficient spatial analysis capabilities and reliable retrieval results even when faced with complex commercial network distributions. In particular, through multi-dimensional spatial relationship analysis, the accuracy and service efficiency of commercial resource management are significantly improved.

[0115] In one embodiment of the AI-based in-vehicle intelligent commercial promotion method of this application, it may further include the following:

[0116] Step S601: Obtain real-time GPS positioning coordinates from the vehicle terminal, retrieve commercial outlets within the current location radius R in the hierarchical commercial information index table, extract the retrieved commercial outlet identification code, query the corresponding commercial information content from the commercial information resource database based on the commercial outlet identification code, calculate the shortest path distance from the vehicle to each commercial outlet based on the outlet adjacency graph, and use the shortest path distance as the location relevance feature.

[0117] Step S602: Input the commercial information content into a collaborative filtering model based on user historical behavior to calculate a collaborative recommendation score, input the commercial information content into a content filtering model based on product attributes to calculate a content recommendation score, and combine the location relevance features to weight and fuse the collaborative recommendation score and the content recommendation score to generate a hybrid recommendation score.

[0118] Optionally, this embodiment addresses the problems of inaccurate location association, single recommendation model, and simplistic fusion strategy in traditional commercial recommendation systems by innovatively designing an intelligent recommendation scheme based on multi-model fusion. This embodiment first acquires GPS positioning information in real time through an in-vehicle terminal and uses spatial indexing technology to perform a range query in a hierarchical commercial information index table: Search_Region = Circle(Position, R), where Position is the current location coordinates and R is the search radius. The system dynamically adjusts the search radius according to road level to ensure that an appropriate number of candidate points are obtained in different scenarios.

[0119] This embodiment deeply optimizes the path analysis mechanism. Based on the network adjacency graph, an improved Dijkstra algorithm is used to calculate the actual driving distance from vehicles to each commercial network point. The system considers factors such as road grade and traffic conditions to construct a weighted path calculation model: Path_Weight = Distance + Traffic_Factor + Road_Level_Weight, where Distance is the geometric distance, Traffic_Factor is the traffic impact factor, and Road_Level_Weight is the road grade weight. The path planning engine optimizes computational efficiency, enabling rapid distance assessment for large-scale network points.

[0120] This embodiment innovatively implements a collaborative filtering strategy. It employs a deep neural network-based collaborative filtering model that integrates user historical behavior data and contextual features. The system captures dynamic changes in user interests through an attention mechanism, paying particular attention to the impact of time and location on behavioral patterns. For example, for users who frequently operate in specific areas, the model strengthens the behavioral references of similar users within that area; when exploring new areas, it focuses more on recommendations from users across the entire user base.

[0121] This embodiment optimizes the content filtering mechanism. Content feature vectors are constructed based on product attributes, and latent feature representations of products are extracted using a deep learning network. The system designs a multi-level feature matching strategy, considering multiple dimensions such as product category, price range, and service characteristics. The degree of content matching is evaluated by calculating the similarity between user profiles and product features. Particularly for new products, the cold start problem is addressed through transfer learning.

[0122] This embodiment deeply optimizes the feature fusion mechanism. An adaptive weight adjustment strategy is designed to dynamically balance the importance of location relevance, collaborative recommendation, and content matching according to different scenarios. For example, in familiar areas, the weight of collaborative recommendation is increased; when exploring new areas, the influence of location relevance is increased. The system continuously optimizes the fusion parameters through online learning methods to adapt to dynamic changes in user preferences.

[0123] This embodiment innovatively implements a hybrid recommendation scoring mechanism. A comprehensive scoring function is constructed: Score = αCollaborative + βContent + γ × Location, where each parameter represents the collaborative recommendation score, content recommendation score, and location relevance score, respectively, and α, β, and γ are the corresponding weight coefficients. The system uses a multi-objective optimization algorithm to solve for the optimal weight combination, balancing recommendation accuracy and diversity. A feedback mechanism continuously optimizes the scoring model to improve recommendation performance.

[0124] This embodiment achieves accurate recommendation of commercial information through multi-model fusion technology. It demonstrates particularly strong pattern recognition capabilities and fusion effects when handling location-related recommendation scenarios. By applying deep learning models, the system can accurately capture the correlation between user preferences and location features, providing reliable decision support for accurate push notifications.

[0125] This embodiment's innovative design not only solves the location association and model fusion problems in traditional methods but also establishes a continuously optimizing recommendation framework. Through continuous optimization of recommendation strategies and improvement of fusion methods, the system can continuously enhance its understanding of user needs, providing strong support for intelligent push notifications. This multi-model fusion-based recommendation mechanism ensures that the system maintains efficient analytical capabilities and reliable recommendation results even in complex business scenarios. In particular, the deep integration of location features significantly improves recommendation accuracy and user experience.

[0126] In one embodiment of the AI-based in-vehicle intelligent commercial promotion method of this application, it may further include the following:

[0127] Step S701: Construct a comprehensive ranking scoring function, assign weight coefficients to the hybrid recommendation score, the scene suitability score, and the click probability prediction value, solve for the optimal weight combination based on a multi-objective optimization algorithm, calculate the comprehensive score of the commercial information to be pushed using the optimal weight combination, normalize the comprehensive score to obtain the ranking score, and sort the commercial information to be pushed from high to low according to the ranking score.

[0128] Step S702: Select the N highest-scoring business information items from the sorting results, extract the titles, description texts, and image materials of the N business information items, adaptively scale the image materials according to the resolution of the vehicle terminal display screen, assemble the titles, description texts, and scaled images into push content according to a preset layout template, and send the push content to the vehicle terminal display screen through the wireless communication network.

[0129] Optionally, this embodiment addresses the problems of simplistic ranking strategies, poor display effects, and subpar user experience inherent in traditional commercial push notifications by innovatively designing an intelligent ranking and display scheme based on multi-objective optimization. This embodiment first constructs a comprehensive scoring function: Score = w1Hybrid + w2Scene + w3 × Click, where Hybrid is the hybrid recommendation score, Scene is the scene suitability score, Click is the predicted click probability, and w1, w2, and w3 are corresponding weight coefficients. The system solves the weight optimization problem using a genetic algorithm to ensure an optimal balance between recommendation relevance, scene suitability, and user interest.

[0130] This embodiment deeply optimizes the multi-objective optimization mechanism. An improved NSGA-II algorithm is used to handle the weight optimization problem, and the optimal solution is found through Pareto front analysis. The system designs adaptive crossover and mutation operators to improve the algorithm's convergence efficiency. Special attention is paid to the dynamic adjustment of weights, dynamically balancing the importance of each objective according to the characteristics of different time periods and scenarios. For example, during morning and evening rush hours, the weight of scenario suitability is increased; during leisure time, more emphasis is placed on recommendation relevance.

[0131] This embodiment innovatively implements a ranking normalization strategy. Based on the optimal weight combination, a comprehensive score for business information is calculated, and the ranking score is processed using a min-max normalization method: Normalized_Score = (Score - Min_Score) / (Max_Score - Min_Score). The system dynamically updates the normalization parameters using sliding window technology to ensure the stability of the ranking results. For special scenarios, such as new product promotions or limited-time events, the normalization score is adjusted and corrected through rule changes to ensure effective push notifications.

[0132] This embodiment optimizes the content filtering mechanism. It selects the N highest-scoring messages from the sorting results, with the value of N dynamically adjusted based on the current scenario and user status. The system considers information diversity and ensures the distinctiveness of the selected content through a maximum marginal relevance algorithm. For continuous push scenarios, sequence pattern analysis avoids content duplication, improving user experience. Especially in driving scenarios, the system controls the amount of information to ensure it does not affect driving safety.

[0133] This embodiment deeply optimizes the content display mechanism. Based on the display characteristics of in-vehicle terminals, an adaptive layout system is designed. Employing responsive design principles, the content layout is dynamically adjusted according to screen resolution. The system intelligently crops and compresses materials using image processing algorithms: Scale_Ratio = min(Screen_Width / Image_Width, Screen_Height / Image_Height), where each parameter represents the width and height of the screen and the image, respectively. A quality assessment model ensures image clarity while optimizing transmission efficiency.

[0134] This embodiment innovatively implements a push control strategy. A complete content delivery control system is constructed, adjusting display duration and update frequency based on vehicle speed and driving status. The system employs a hierarchical display strategy, using different display methods for important and ordinary information. An attention prediction model is used to assess the visual salience of content and optimize the placement of key information. Especially in complex road conditions, the system automatically simplifies the displayed content to ensure driving safety.

[0135] This embodiment achieves intelligent sorting and display of commercial information through multi-objective optimization technology. It demonstrates particularly strong decision-making capabilities and display effects when handling complex push scenarios. Through the application of deep learning models, the system can accurately grasp user needs, rationally arrange push content, and provide users with a high-quality information service experience.

[0136] This innovative design not only solves the sorting effectiveness and display experience problems of traditional methods, but also establishes a continuously optimizeable push framework. Through continuous optimization of the sorting strategy and improvement of the display method, the system can continuously improve its responsiveness to user needs, providing strong support for intelligent push notifications. This push mechanism based on multi-objective optimization ensures that the system maintains efficient sorting capabilities and reliable display effects when facing dynamically changing scenarios. In particular, the application of adaptive display technology significantly improves the user experience and conversion rate of commercial information push notifications.

[0137] To effectively address the shortcomings of traditional technologies in areas such as user profiling, push timing, and content recommendation, and to significantly improve the intelligence level and conversion rate of in-vehicle commercial promotion, this application provides an embodiment of an AI-based in-vehicle intelligent commercial promotion device for implementing all or part of the aforementioned AI-based in-vehicle intelligent commercial promotion method. See [link to embodiment]. Figure 2 The AI-based in-vehicle intelligent business promotion device specifically includes the following components:

[0138] User profile construction module 10 is used to collect positioning data, navigation trajectory data and vehicle sensor data from vehicle terminals, perform sliding time window analysis on the vehicle sensor data to generate environmental feature vectors, extract driving behavior features from the positioning data and navigation trajectory data, construct a driving behavior feature model, perform driving habit clustering analysis on the user based on the driving behavior feature model, construct a spatiotemporal behavior pattern recognition model in combination with the environmental feature vectors, write the output of the spatiotemporal behavior pattern recognition model into the user profile database, and train a user interest prediction model based on the user profile database. The user interest prediction model includes an interest decay coefficient.

[0139] The commercial service determination module 20 is used to read the advertising content and merchant service information in the commercial information resource database, extract the feature vector of commercial information, obtain external environment data, input the commercial information feature vector and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push timing in different scenarios, construct a commercial resource distribution map based on the geographic information system, mark commercial outlets in layers according to road type, and generate a layered commercial information index table.

[0140] The intelligent recommendation module 30 is used to obtain the real-time positioning coordinates of the vehicle, match the commercial information around the current location in the commercial resource distribution map, input the commercial information into the collaborative filtering model and the content filtering model to calculate the hybrid recommendation score, input the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into the comprehensive ranking model, the comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and send the push content to the vehicle terminal display screen.

[0141] As described above, the AI-based in-vehicle intelligent commercial promotion device provided in this application embodiment can achieve accurate identification of spatiotemporal behavioral patterns by innovatively constructing a user profile mechanism and integrating location data, navigation trajectory, and sensor data. It designs a scenario-based push strategy, combining external environmental data and commercial resource distribution to establish a hierarchical commercial information index for intelligent matching. A multi-model fusion mechanism is introduced, using collaborative filtering, content filtering, and a comprehensive ranking model to achieve multi-objective optimization of personalized push content. This method effectively solves the shortcomings of traditional technologies in user profiling, push timing, and content recommendation, significantly improving the intelligence level and conversion effect of in-vehicle commercial promotion.

[0142] From a hardware perspective, in order to effectively address the shortcomings of traditional technologies in user profiling, push timing, and content recommendation, and significantly improve the intelligence level and conversion effect of in-vehicle commercial promotion, this application provides an embodiment of an electronic device for implementing all or part of the AI-based in-vehicle intelligent commercial promotion method. The electronic device specifically includes the following components:

[0143] The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the AI-based in-vehicle intelligent business promotion device and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the AI-based in-vehicle intelligent business promotion method and the embodiments of the AI-based in-vehicle intelligent business promotion device, the content of which is incorporated herein, and repeated details will not be described again.

[0144] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0145] In practical applications, some parts of the AI-based in-vehicle intelligent business promotion method can be executed on the electronic device side as described above, or all operations can be completed in the client device. The specific choice depends on the processing power of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed in the client device, the client device may further include a processor.

[0146] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0147] Figure 3 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 3 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 3 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0148] In one embodiment, the AI-based in-vehicle intelligent business promotion method can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following controls:

[0149] Step S101: Collect positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal; perform sliding time window analysis on the vehicle sensor data to generate an environmental feature vector; extract driving behavior features from the positioning data and the navigation trajectory data; construct a driving behavior feature model; perform driving habit clustering analysis on the user based on the driving behavior feature model; construct a spatiotemporal behavior pattern recognition model by combining the environmental feature vector; write the output of the spatiotemporal behavior pattern recognition model into a user profile database; train a user interest prediction model based on the user profile database; the user interest prediction model includes an interest decay coefficient.

[0150] Step S102: Read the advertising content and merchant service information in the commercial information resource database, extract the feature vector of the commercial information, obtain external environment data, input the commercial information feature vector and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push time in different scenarios, construct a commercial resource distribution map based on the geographic information system, mark the commercial outlets in layers according to road type, and generate a layered commercial information index table.

[0151] Step S103: Obtain the real-time location coordinates of the vehicle, match the commercial information around the current location in the commercial resource distribution map, input the commercial information into the collaborative filtering model and the content filtering model to calculate the hybrid recommendation score, input the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into the comprehensive ranking model, the comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and send the push content to the vehicle terminal display screen.

[0152] As described above, the electronic device provided in this application innovatively constructs a user profiling mechanism, integrating location data, navigation trajectory, and sensor data to achieve accurate identification of spatiotemporal behavioral patterns. It designs a scenario-based push strategy, combining external environmental data and commercial resource distribution to establish a hierarchical commercial information index for intelligent matching. A multi-model fusion mechanism is introduced, employing collaborative filtering, content filtering, and a comprehensive ranking model to achieve multi-objective optimization of personalized push content. This method effectively addresses the shortcomings of traditional technologies in user profiling, push timing, and content recommendation, significantly improving the intelligence level and conversion rate of in-vehicle commercial promotion.

[0153] In another embodiment, the AI-based in-vehicle intelligent business promotion device can be configured separately from the central processing unit 9100. For example, the AI-based in-vehicle intelligent business promotion device can be configured as a chip connected to the central processing unit 9100, and the functions of the AI-based in-vehicle intelligent business promotion method can be realized through the control of the central processing unit.

[0154] like Figure 3 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 3 All components shown; in addition, the electronic device 9600 may also include Figure 3 For components not shown, please refer to existing technologies.

[0155] like Figure 3 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0156] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0157] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0158] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0159] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device's communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0160] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0161] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored audio via the speaker 9131.

[0162] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the AI-based in-vehicle intelligent business promotion method with a server or client as the execution subject in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the AI-based in-vehicle intelligent business promotion method with a server or client as the execution subject in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0163] Step S101: Collect positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal; perform sliding time window analysis on the vehicle sensor data to generate an environmental feature vector; extract driving behavior features from the positioning data and the navigation trajectory data; construct a driving behavior feature model; perform driving habit clustering analysis on the user based on the driving behavior feature model; construct a spatiotemporal behavior pattern recognition model by combining the environmental feature vector; write the output of the spatiotemporal behavior pattern recognition model into a user profile database; train a user interest prediction model based on the user profile database; the user interest prediction model includes an interest decay coefficient.

[0164] Step S102: Read the advertising content and merchant service information in the commercial information resource database, extract the feature vector of the commercial information, obtain external environment data, input the commercial information feature vector and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push time in different scenarios, construct a commercial resource distribution map based on the geographic information system, mark the commercial outlets in layers according to road type, and generate a layered commercial information index table.

[0165] Step S103: Obtain the real-time location coordinates of the vehicle, match the commercial information around the current location in the commercial resource distribution map, input the commercial information into the collaborative filtering model and the content filtering model to calculate the hybrid recommendation score, input the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into the comprehensive ranking model, the comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and send the push content to the vehicle terminal display screen.

[0166] As described above, the computer-readable storage medium provided in this application innovatively constructs a user profiling mechanism, integrating location data, navigation trajectories, and sensor data to achieve accurate identification of spatiotemporal behavioral patterns. It designs a scenario-based push strategy, combining external environmental data and commercial resource distribution to establish a hierarchical commercial information index for intelligent matching. A multi-model fusion mechanism is introduced, employing collaborative filtering, content filtering, and a comprehensive ranking model to achieve multi-objective optimization of personalized push content. This method effectively addresses the shortcomings of traditional technologies in user profiling, push timing, and content recommendation, significantly improving the intelligence level and conversion rate of in-vehicle commercial promotion.

[0167] Embodiments of this application also provide a computer program product capable of implementing all steps of the AI-based in-vehicle intelligent business promotion method in the above embodiments, where the execution subject is a server or a client. When executed by a processor, this computer program / instruction implements the steps of the AI-based in-vehicle intelligent business promotion method. For example, the computer program / instruction implements the following steps:

[0168] Step S101: Collect positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal; perform sliding time window analysis on the vehicle sensor data to generate an environmental feature vector; extract driving behavior features from the positioning data and the navigation trajectory data; construct a driving behavior feature model; perform driving habit clustering analysis on the user based on the driving behavior feature model; construct a spatiotemporal behavior pattern recognition model by combining the environmental feature vector; write the output of the spatiotemporal behavior pattern recognition model into a user profile database; train a user interest prediction model based on the user profile database; the user interest prediction model includes an interest decay coefficient.

[0169] Step S102: Read the advertising content and merchant service information in the commercial information resource database, extract the feature vector of the commercial information, obtain external environment data, input the commercial information feature vector and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push time in different scenarios, construct a commercial resource distribution map based on the geographic information system, mark the commercial outlets in layers according to road type, and generate a layered commercial information index table.

[0170] Step S103: Obtain the real-time location coordinates of the vehicle, match the commercial information around the current location in the commercial resource distribution map, input the commercial information into the collaborative filtering model and the content filtering model to calculate the hybrid recommendation score, input the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into the comprehensive ranking model, the comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and send the push content to the vehicle terminal display screen.

[0171] As described above, the computer program product provided in this application innovatively constructs a user profiling mechanism, integrating location data, navigation trajectories, and sensor data to achieve accurate identification of spatiotemporal behavioral patterns. It designs a scenario-based push strategy, combining external environmental data and commercial resource distribution to establish a hierarchical commercial information index for intelligent matching. A multi-model fusion mechanism is introduced, employing collaborative filtering, content filtering, and a comprehensive ranking model to achieve multi-objective optimization of personalized push content. This method effectively addresses the shortcomings of traditional technologies in user profiling, push timing, and content recommendation, significantly improving the intelligence level and conversion rate of in-vehicle commercial promotion.

[0172] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0173] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0174] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0175] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0176] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. An AI-based in-vehicle intelligent business promotion method, characterized in that, The method includes: The system collects positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal. It performs sliding time window analysis on the vehicle sensor data to generate an environmental feature vector. Driving behavior features are extracted from the positioning data and navigation trajectory data to construct a driving behavior feature model. Based on the driving behavior feature model, it performs driving habit clustering analysis on the user. Combined with the environmental feature vector, it constructs a spatiotemporal behavior pattern recognition model. The output of the spatiotemporal behavior pattern recognition model is written into a user profile database. Based on the user profile database, it trains a user interest prediction model, which includes an interest decay coefficient. Read advertising content and merchant service information from the commercial information resource database, extract feature vectors of commercial information, obtain external environment data, input the commercial information feature vectors and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push timing in different scenarios, construct a commercial resource distribution map based on the geographic information system, mark commercial outlets in layers according to road type, and generate a layered commercial information index table. The system obtains the real-time location coordinates of the vehicle, matches the surrounding commercial information in the commercial resource distribution map, inputs the commercial information into a collaborative filtering model and a content filtering model to calculate a hybrid recommendation score, inputs the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into a comprehensive ranking model, the comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and sends the push content to the vehicle terminal display screen.

2. The AI-based in-vehicle intelligent business promotion method according to claim 1, characterized in that, The process involves collecting positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal; performing sliding time window analysis on the vehicle sensor data to generate environmental feature vectors; extracting driving behavior features from the positioning data and navigation trajectory data; and constructing a driving behavior feature model, including: The vehicle terminal reads real-time coordinate information from the GPS positioning module, path planning data from the navigation system, vehicle speed data from the vehicle CAN bus, and temperature and light data from the vehicle environment sensor. Based on the data acquisition timestamp, the data from different sources is time-series aligned, and the aligned data is stored in the terminal cache database. The data preprocessing module is used to normalize the cached data and remove outliers. Feature extraction is performed on the vehicle environment sensor data using a sliding time window of length N. Statistical features of the data within the window are calculated to generate an environmental feature vector. The vehicle's position trajectory and speed change curve are extracted from the positioning data. The vehicle's path selection preference and driving time distribution are extracted from the navigation trajectory data. The environmental feature vector, along with the position trajectory, speed change curve, path selection preference, and driving time distribution, are input into a driving behavior feature extraction model to generate a user driving behavior feature vector.

3. The AI-based in-vehicle intelligent business promotion method according to claim 1, characterized in that, The process involves clustering user driving habits based on the driving behavior feature model, constructing a spatiotemporal behavior pattern recognition model by combining the environmental feature vectors, writing the output of the spatiotemporal behavior pattern recognition model into a user profile database, and training a user interest prediction model based on the user profile database. The user interest prediction model includes an interest decay coefficient, comprising: The user's driving behavior feature vector is input into the K-means clustering algorithm. The user is clustered by calculating the Euclidean distance between the feature vectors, and driving habit clustering labels are generated. The environmental feature vector and the driving habit clustering labels are combined to construct a sample dataset. A spatiotemporal behavior pattern recognition model is trained using a long short-term memory neural network. The spatiotemporal behavior pattern recognition model outputs the user's behavior type label and the probability of behavior occurrence. The behavior type label, the probability of occurrence of the behavior, and the user's basic attribute information are written into the user profile database. A commercial information preference matrix is ​​constructed based on the user's historical click behavior data. A time decay function is introduced to calculate the timeliness weight of historical behaviors. The historical behavior data in the commercial information preference matrix is ​​weighted according to the timeliness weight. A neural collaborative filtering algorithm is used to train a user interest prediction model. The user interest prediction model outputs the predicted click probability values ​​of users for different types of commercial information.

4. The AI-based in-vehicle intelligent business promotion method according to claim 1, characterized in that, The process involves reading advertising content and merchant service information from a business information resource database, extracting feature vectors from the business information, acquiring external environmental data, and inputting the business information feature vectors and the external environmental data into a scene recognition model. This scene recognition model is used to identify the optimal push timing for different scenarios, including: The advertising content and merchant service information are read from the business information resource database. The advertising text content is segmented into words. Based on the word vector model, the segmentation results are converted into text semantic vectors. The visual features of the advertising images are extracted. The geographical coordinates, business hours and service type tags of the merchant services are extracted. The text semantic vector, the visual features, the geographical coordinates, the business hours and the service type tags are combined to construct a business information feature vector. Weather data is obtained from a meteorological data interface, road congestion index is obtained from a traffic management platform, and holiday information is obtained from a calendar service. The weather data, road congestion index, and holiday information are used to construct an external environment feature vector. The commercial information feature vector and the external environment feature vector are input into a scene recognition model trained based on a deep neural network. The scene recognition model outputs a suitability score for pushing various types of commercial information in the current scene.

5. The AI-based in-vehicle intelligent business promotion method according to claim 1, characterized in that, The method of constructing a commercial resource distribution map based on a geographic information system, stratifying commercial outlets according to road type, and generating a stratified commercial information index table includes: The system calls the geographic information system interface to obtain electronic map data, extracts the road network topology, classifies and encodes roads according to highways, urban arterial roads, secondary arterial roads and branch roads, reads the geographical coordinates of commercial outlets, calculates the distance from each commercial outlet to the nearest road node, establishes spatial association between commercial outlets and the nearest road node, and marks the location and coverage of commercial outlets on the electronic map based on the spatial association. Commercial outlets are stored in a hierarchical manner. Commercial outlets at the same road level are constructed into an outlet set. The distance matrix between each commercial outlet within the outlet set is calculated. An adjacency graph of outlets is constructed based on the distance matrix. A unique identifier is assigned to each commercial outlet. The commercial outlet identifier, road level, location coordinates, coverage area, and adjacency relationship are written into the hierarchical commercial information index table.

6. The AI-based in-vehicle intelligent commercial promotion method according to claim 1, characterized in that, The process of obtaining the vehicle's real-time location coordinates, matching surrounding commercial information in the commercial resource distribution map, and inputting the commercial information into a collaborative filtering model and a content filtering model to calculate a hybrid recommendation score includes: The vehicle terminal obtains real-time GPS positioning coordinates, retrieves commercial outlets within the current location radius R in the hierarchical commercial information index table, extracts the retrieved commercial outlet identification code, queries the corresponding commercial information content from the commercial information resource database based on the commercial outlet identification code, calculates the shortest path distance from the vehicle to each commercial outlet based on the outlet adjacency graph, and uses the shortest path distance as a location relevance feature. The commercial information content is input into a collaborative filtering model based on user historical behavior to calculate a collaborative recommendation score, and the commercial information content is input into a content filtering model based on product attributes to calculate a content recommendation score. The collaborative recommendation score and the content recommendation score are then weighted and fused together with the location relevance features to generate a hybrid recommendation score.

7. The AI-based in-vehicle intelligent commercial promotion method according to claim 1, characterized in that, The process involves inputting the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into a comprehensive ranking model. This comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and then sends the push content to the vehicle terminal display screen. The process includes: A comprehensive ranking and scoring function is constructed, and the hybrid recommendation score, the scene suitability score, and the click probability prediction value are assigned weight coefficients respectively. The optimal weight combination is solved based on a multi-objective optimization algorithm. The comprehensive score of the commercial information to be pushed is calculated using the optimal weight combination. The comprehensive score is normalized to obtain the ranking score. The commercial information to be pushed is sorted from high to low according to the ranking score. Select the N highest-scoring business information items from the sorting results, extract the titles, descriptions, and images of the N business information items, adaptively scale the images according to the resolution of the vehicle terminal display screen, assemble the titles, descriptions, and scaled images into push content according to a preset layout template, and send the push content to the vehicle terminal display screen through a wireless communication network.

8. An AI-based in-vehicle intelligent commercial promotion device, characterized in that, The device includes: The user profile construction module is used to collect positioning data, navigation trajectory data, and vehicle sensor data from the vehicle terminal; perform sliding time window analysis on the vehicle sensor data to generate environmental feature vectors; extract driving behavior features from the positioning data and navigation trajectory data; construct a driving behavior feature model; perform driving habit clustering analysis on the user based on the driving behavior feature model; construct a spatiotemporal behavior pattern recognition model by combining the environmental feature vectors; write the output of the spatiotemporal behavior pattern recognition model into the user profile database; and train a user interest prediction model based on the user profile database, wherein the user interest prediction model includes an interest decay coefficient. The business service determination module is used to read the advertising content and merchant service information in the business information resource database, extract the feature vector of the business information, obtain external environment data, input the business information feature vector and the external environment data into the scene recognition model, the scene recognition model is used to identify the optimal push timing in different scenarios, construct a business resource distribution map based on the geographic information system, mark the business outlets in layers according to road type, and generate a layered business information index table. The intelligent recommendation module is used to obtain the real-time location coordinates of the vehicle, match the commercial information around the current location in the commercial resource distribution map, input the commercial information into the collaborative filtering model and the content filtering model to calculate the hybrid recommendation score, and input the hybrid recommendation score, the push timing output by the scene recognition model, and the click probability output by the user interest prediction model into the comprehensive ranking model. The comprehensive ranking model uses a multi-objective optimization algorithm to generate push content, and the push content is sent to the vehicle terminal display screen.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the AI-based in-vehicle intelligent business promotion method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the AI-based in-vehicle intelligent business promotion method according to any one of claims 1 to 7.

Images