Service place recommendation method and device, vehicle and storage medium

Abstract

The invention provides a service place recommendation method and device, a vehicle and a storage medium, and is applied to the technical field of Internet of Vehicles. The method comprises the following steps: acquiring original input data including a vehicle condition, a user behavior and an environment context; determining a current driving scene according to the original input data; judging whether the current driving scene belongs to a preset driving scene set or not to obtain a judgment result; a target recommendation strategy is determined from a first recommendation strategy and a second recommendation strategy according to the judgment result, the first recommendation strategy is based on a reinforcement learning algorithm, and the second recommendation strategy is based on big language model reasoning; and recommending the target service place to the user based on the target recommendation strategy. According to the service place recommendation method and device, the driving scene can be accurately judged based on the multi-modal data, different recommendation strategies are selected according to the judgment result, it is ensured that reliable service place recommendation can be provided under both the known driving scene and the unknown driving scene, and therefore the accuracy of service place recommendation and the user experience are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of vehicle networking technology, specifically to a service location recommendation method, apparatus, vehicle, and computer-readable storage medium. Background Technology

[0002] With the rapid development of vehicle-to-everything (V2X) and smart cockpit technologies, intelligent recommendations for in-vehicle services have become crucial for enhancing user experience. However, traditional solutions typically employ a uniform recommendation logic across all driving scenarios. On one hand, relying solely on vehicle status data or user historical behavior data for service delivery makes it difficult to accurately identify the user's true intent during complex and dynamic driving processes, leading to discrepancies between recommended results and actual needs. On the other hand, especially in new scenarios lacking sufficient historical interaction data, such as with new users, new services, or unexpected situations, this single mechanism often fails to effectively activate adaptation strategies, making it difficult to generate relevant service suggestions and potentially even outputting mismatched service recommendations, thus severely impacting the accuracy and reliability of service recommendations. Summary of the Invention

[0003] This application provides a service location recommendation method, apparatus, vehicle, and computer-readable storage medium. This application can accurately identify driving scenarios based on multimodal data and select different recommendation strategies according to the identification results, ensuring reliable service location recommendations in both known and unknown driving scenarios, thereby significantly improving the accuracy of service location recommendations and user experience.

[0004] Firstly, a service location recommendation method is provided. This method includes: acquiring raw input data, including vehicle condition data, user behavior data, and environmental context data; determining the current driving scenario based on the raw input data; determining whether the current driving scenario belongs to a preset set of driving scenarios, and obtaining a judgment result; determining a target recommendation strategy from a first recommendation strategy and a second recommendation strategy based on the judgment result, wherein the first recommendation strategy is a recommendation strategy based on a reinforcement learning algorithm, and the second recommendation strategy is a recommendation strategy based on large language model reasoning; and recommending target service locations to the user based on the target recommendation strategy.

[0005] Based on the above technical solution, this application embodiment achieves accurate judgment of the current driving scenario by collecting multimodal data of vehicle condition, user behavior and environmental context, and selects different recommendation strategies according to the judgment results. This avoids the problem that traditional methods rely on only a single recommendation strategy and are difficult to effectively recommend service locations when facing unknown driving scenarios. It ensures that reliable service location recommendations can be provided in both known and unknown driving scenarios, thereby significantly improving the accuracy of service location recommendations and user experience.

[0006] In conjunction with the first aspect, in some implementations of the first aspect, the above-mentioned determination of the current driving scenario based on the original input data includes: extracting features from the original input data to obtain original data features; performing chi-square filtering on the original data features to obtain first filtered features; performing variance filtering on the first filtered features to obtain second filtered features; performing noise reduction and data reconstruction on the second filtered features to obtain core data features; and determining the current driving scenario based on the core data features.

[0007] Based on the above technical solution, this application embodiment performs a hierarchical processing flow of feature extraction, chi-square filtering, variance filtering, and noise reduction reconstruction sequentially on the multimodal raw input data. This extracts core data features that can accurately characterize the essential attributes of driving scenarios. Based on these core data features, the current driving scenario is determined, which effectively improves the accuracy and reliability of current driving scenario recognition. This provides an accurate and solid basis for the adaptive selection of subsequent recommendation strategies, avoids recommendation strategy matching errors caused by scenario recognition bias, and ensures the effectiveness and stability of the entire service location recommendation process.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, determining the target recommendation strategy from the first recommendation strategy and the second recommendation strategy based on the judgment result includes: when the judgment result indicates that the current driving scenario belongs to a preset driving scenario set, determining the first recommendation strategy as the target recommendation strategy; when the judgment result indicates that the current driving scenario does not belong to the preset driving scenario set, determining the second recommendation strategy as the target recommendation strategy.

[0009] Based on the above technical solution, this application embodiment dynamically selects different recommendation methods according to the judgment result: in known driving scenarios, a recommendation method based on reinforcement learning is adopted; in unknown driving scenarios, a reasoning method based on a large language model is switched, realizing dynamic adaptation of recommendation strategies and scenario states. This ensures the accuracy and personalization of recommendation decisions in known driving scenarios and improves the effectiveness of recommendation services in unknown driving scenarios, providing a reliable basis for subsequent recommendation of service locations and effectively improving the accuracy of service location recommendations and user experience.

[0010] In conjunction with the first aspect, in certain implementations of the first aspect, when the target recommendation strategy is the first recommendation strategy, the above-mentioned recommendation of target service locations to users based on the target recommendation strategy includes: determining a set of services to be recommended based on the current driving scenario, the set of services to be recommended including multiple candidate service locations; obtaining preset scores corresponding to multiple candidate service locations, the preset scores being the user's past experience ratings for candidate service locations; using the current driving scenario, multiple candidate service locations, and the preset scores corresponding to multiple candidate service locations as the scenario space state, immediate reward, and recommendation action in the reinforcement learning algorithm, respectively, to obtain the recommendation scores corresponding to each of the multiple candidate service locations; and determining the target service location based on the recommendation scores corresponding to each of the multiple candidate service locations and the user's past preference for the multiple candidate service locations.

[0011] Based on the above technical solution, this application integrates driving scenarios, candidate service locations, and historical experience scores into a reinforcement learning algorithm to calculate the recommendation score for each candidate service location. Then, it combines this with the user's past preferences for each candidate service location to determine the target service location. This allows the algorithm to automatically learn service preferences from the user's actual historical interactions without relying on preset rules. This approach fully utilizes the user's past usage preferences and service feedback, while incorporating user feedback into the recommendation score calculation process. This makes service recommendations in known driving scenarios more aligned with the user's personalized needs, effectively improving the alignment between the recommendation results and the user's actual needs, thereby ensuring the accuracy of service location recommendations.

[0012] In combination with the first aspect and the above implementation methods, in some implementation methods of the first aspect, determining the target service location based on the recommendation scores corresponding to each of the multiple candidate service locations and the user's past preference for the multiple candidate service locations includes: determining the candidate service location with the highest score from the recommendation scores corresponding to each of the multiple candidate service locations to obtain the first service location; and determining the first service location as the target service location if the preference for the first service location is greater than a preset preference threshold.

[0013] Based on the above technical solution, this application embodiment employs a two-stage screening and judgment process to ensure that the service location recommendation results are suitable for the current driving scenario and align with users' long-term usage habits. First, the highest-scoring service location from the candidate service location recommendation scores output by the reinforcement learning algorithm is selected as the first service location, ensuring the recommendation result matches the current context. Then, the user's historical preference level for the first service location is compared with a preset preference threshold. Only when the preference level of the first service location exceeds this threshold is it determined as the target service location. This effectively avoids the problem of relying solely on algorithmic scoring for recommendations while ignoring actual user habits. The recommendation results not only meet scenario adaptation standards but also closely reflect users' real needs and preferences, thereby significantly improving the accuracy of service location recommendations in known driving scenarios and optimizing the user's actual experience.

[0014] In conjunction with the first aspect, in some implementations of the first aspect, when the target recommendation strategy is the second recommendation strategy, the above-mentioned recommendation of target service locations for users based on the target recommendation strategy includes: converting the current driving scenario into structured text prompt information; inputting the text prompt information into a large language model, and having the large language model output user needs; determining a preset service location that meets the user needs from multiple preset service locations to obtain a second service location; and determining the target service location based on the second service location.

[0015] Based on the above technical solution, this application embodiment, when facing unknown or newly emerging driving scenarios, no longer relies on user historical data. Instead, it converts the current situation into structured text prompts and inputs them into a large language model. Based on the large language model, it understands and infers the services the user may need, and then selects candidate service locations that meet the needs from preset service locations. This ensures that reasonable recommendations can be given even without historical interaction records, avoiding the problems of traditional recommendation methods that "cannot recommend" or "recommend randomly" in new users, new routes, or unexpected situations. It effectively improves the accuracy of service location recommendations and user experience in new scenarios.

[0016] In conjunction with the first aspect and the above implementation methods, in some implementation methods of the first aspect, the above-mentioned determination of the target service location based on the second service location includes: if the second service location is a single location, then the second service location is determined as the target service location; if there are multiple second service locations, then the evaluation score of each second service location is calculated; the second service location with the highest recommended score is determined from the recommended scores of each second service location to obtain the third service location; and the third service location is determined as the target service location.

[0017] Based on the above technical solution, this application embodiment adopts a flexible filtering method for the candidate service locations output by the large language model under the second recommendation strategy. When a single service location is matched, it is directly adopted to ensure recommendation efficiency. When there are multiple candidate service locations that meet the user's needs, the optimal option is selected by calculating the evaluation score. It does not randomly select from multiple candidate service locations, thereby avoiding random or unsatisfactory recommendation results in the case of multiple candidates. This makes the service location recommendation in the new scenario more reliable and further improves the accuracy of the recommendation.

[0018] Secondly, a service venue recommendation device is provided, the service venue recommendation device comprising: The data acquisition module is used to acquire raw input data, which includes vehicle condition data, user behavior data, and environmental context data. The scenario determination module is used to determine the current driving scenario based on the original input data; The scene judgment module is used to determine whether the current driving scene belongs to the preset driving scene set and obtain the judgment result; The strategy determination module is used to determine the target recommendation strategy from the first recommendation strategy and the second recommendation strategy based on the judgment result. The first recommendation strategy is a recommendation strategy based on the reinforcement learning algorithm, and the second recommendation strategy is a recommendation strategy based on the reasoning of the large language model. The service recommendation module is used to recommend target service locations to users based on the target recommendation strategy.

[0019] Thirdly, a vehicle is provided, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, causing the vehicle to perform the service location recommendation method in the first aspect or any possible implementation thereof.

[0020] Fourthly, a computer program product is provided, comprising: computer program code, which, when run on a computer, causes the computer to perform the service location recommendation method in the first aspect or any possible implementation thereof.

[0021] Fifthly, a computer-readable storage medium is provided, which stores a computer program that, when executed, causes the computer to perform the service location recommendation method in the first aspect or any possible implementation thereof. Attached Figure Description

[0022] Figure 1 A flowchart illustrating a service location recommendation method provided in an embodiment of this application is shown; Figure 2 A schematic diagram of the architecture of a service location recommendation method provided in an embodiment of this application is shown; Figure 3 This illustration shows a schematic diagram of a service location recommendation device provided in an embodiment of this application; Figure 4 A schematic diagram of the structure of a vehicle provided in an embodiment of this application is shown. Detailed Implementation

[0023] The technical solutions in this application will be clearly and thoroughly described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. "And / or" in the text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more than two.

[0024] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0025] With the rapid development of vehicle-to-everything (V2X) and smart cockpit technologies, intelligent recommendations for in-vehicle services have become crucial for enhancing user experience. However, traditional solutions typically employ a uniform recommendation logic across all driving scenarios. On one hand, relying solely on vehicle status data or user historical behavior data for service delivery makes it difficult to accurately identify the user's true intent during complex and dynamic driving processes, leading to discrepancies between recommended results and actual needs. On the other hand, especially in new scenarios lacking sufficient historical interaction data, such as with new users, new services, or unexpected situations, this single mechanism often fails to effectively activate adaptation strategies, making it difficult to generate relevant service suggestions and potentially even outputting mismatched service recommendations, thus severely impacting the accuracy and reliability of service recommendations.

[0026] To address the aforementioned issues, this application provides a service location recommendation method, apparatus, vehicle, and computer-readable storage medium. This application can accurately identify driving scenarios based on multimodal data and select different recommendation strategies according to the identification results, ensuring reliable service location recommendations in both known and unknown driving scenarios, thereby significantly improving the accuracy of service location recommendations and user experience.

[0027] Figure 1 This illustration shows a flowchart of a service location recommendation method provided in an embodiment of this application; specifically as follows: Figure 1 As shown, in this embodiment, the executing entity of the method is the vehicle-mounted system, and the method includes the following steps: S110: Obtain raw input data, which includes vehicle condition data, user behavior data, and environmental context data.

[0028] Vehicle condition data includes, but is not limited to, remaining battery power, current fuel consumption, driving range, vehicle speed, continuous driving time, tire pressure, braking frequency, and fatigue driving-related data (such as driver continuous operation time and vehicle driving stability parameters). User behavior data includes, but is not limited to, user history of service usage selection records (such as charging stations, rest areas, audio / video playback preferences), in-vehicle function interaction behavior (voice commands, touch operations, navigation destination setting records), service feedback data (acceptance / rejection of recommended services, ratings and evaluations), and long-term usage habit data. Environmental context data includes, but is not limited to, real-time traffic information (current geographical location, real-time traffic conditions, congestion level, road type, construction section warnings), current regional weather conditions (temperature, precipitation, visibility), points of interest information along the route (service location type, location coordinates, distance, operating status, user reviews), regional attributes (such as business districts, service areas, residential areas), traffic control information, and regional network signal status.

[0029] Specifically, vehicle status data is read in real time from the vehicle controller, battery management system, engine control unit, anti-lock braking system, tire pressure monitoring system, and advanced driver assistance system via the vehicle CAN bus interface; user behavior data is collected through the human-machine interaction log module of the in-vehicle operating system, including user voice command text, touch screen operation trajectory, and destination history of navigation applications, and the service management module records user calls, confirmations, cancellations, and ratings for various in-vehicle services. At the same time, it combines the user profile engine to model long-term usage habits and output structured preference data; environmental context data is obtained through the positioning module (supporting GPS / BeiDou) to obtain the current latitude and longitude and timestamp, through the in-vehicle telematics unit via the cellular network (4G / 5G) to call the application programming interface of a third-party map service provider to obtain real-time traffic conditions, road events, and details of points of interest within the surrounding preset range, through the meteorological service interface to obtain the weather data of the current location, through the electronic map database to parse regional attributes, and through the communication module to report the current cellular network signal strength and connection status.

[0030] S120: Determine the current driving scenario based on the original input data.

[0031] Specifically, the acquired multi-source heterogeneous raw input data is first fused to form a unified multi-dimensional raw feature vector. Then, a multi-stage feature selection and optimization method, employing chi-square filtering, variance filtering, noise reduction filtering, and data reconstruction, is used to extract a core feature vector from the raw feature vector that is highly relevant, highly discriminative, and low-noise. Specifically, chi-square filtering retains features with significant statistical correlation to the service recommendation tags, variance filtering removes redundant features with weak variation and low discriminative ability across different samples, and noise reduction and data reconstruction reduce the dimensionality and remove noise from the features after the two rounds of filtering, achieving optimized feature reorganization. The core feature vector obtained after the above joint filtering and reconstruction can more accurately and stably represent the essential attributes of the current driving state. Finally, based on this core feature vector, combined with a preset scene feature template or dynamic scene recognition model (such as a scene classifier or clustering model), the current driving state of the vehicle, user behavior tendencies, and environmental context conditions are comprehensively assessed to accurately determine the current driving scenario.

[0032] In one possible implementation, determining the current driving scenario based on the original input data includes the following steps: Feature extraction is performed on the original input data to obtain the original data features; The original data features are subjected to chi-square filtering to obtain the first filtered features; The first filtering feature is subjected to variance filtering to obtain the second filtering feature; The second filtered feature is subjected to noise reduction and data reconstruction to obtain the core data feature; The current driving scenario is determined based on core data features.

[0033] Among them, the original data features refer to the high-dimensional feature set obtained after vectorizing, normalizing and structuring the original input data, which contains all quantifiable initial information extracted from multi-source data; the first filtering features refer to the feature subset retained after chi-square filtering of the original data feature set; the second filtering features refer to the feature subset retained after variance filtering of the first filtering features; and the core data features refer to the feature subset retained after noise reduction and data reconstruction of the second filtering features.

[0034] Specifically, the in-vehicle system first performs multimodal fusion processing on the acquired vehicle condition data, user behavior data, and environmental context data. Heterogeneous data from different sources are vectorized (e.g., numerical data such as remaining battery power and driving range are directly quantified, user "accept / reject" operations are converted into binary identifiers, and environmental context "congestion level" is converted into discrete values ​​of 1-5). Then, all vectorized data is normalized (uniformly mapped to the [0,1] interval). Finally, the vectorized results of the three types of data are concatenated according to preset dimensions into a structured high-dimensional feature set, i.e., the original data features, whose corresponding feature vector is Vraw=[f1,f2,f3,...,fm], where m is the original data feature dimension. Let represent the i-th feature value. This high-dimensional feature vector encompasses all quantifiable initial information in the original input data.

[0035] Building upon this foundation, the in-vehicle system employs a combination of chi-square filtering, variance filtering, and a noise-reducing autoencoder to extract core data features from the original data, thereby eliminating data redundancy and improving processing efficiency. First, chi-square filtering is performed, using the chi-square test algorithm to select features most relevant to the driving scenario determination labels. This is achieved by calculating the correlation between each feature and the data. Chi-square statistics The chi-square statistic, used to measure the degree of association between a statistic and scene labels, is calculated using the following formula:

[0036] in, It is a feature The observation frequency in the j-th scenario It is a feature Expected frequency in the j-th scenario; filter out those with chi-square statistics values ​​higher than a preset threshold. The features are used to form the feature subset corresponding to the first filtering feature.

[0037] Next, variance filtering is performed, calculating the variance of each feature in the first filtering feature set. variance To eliminate redundant features with weak differences and low discrimination among samples, the variance is calculated using the following formula:

[0038] in, It is a feature The mean of all samples, It is a feature The variance of the selected samples in the j-th sample is higher than the preset threshold. The features are used to form the feature subset corresponding to the second filtering feature.

[0039] Finally, noise reduction and data reconstruction are performed. A noise reduction autoencoder is used to reduce the dimensionality and remove noise from the second filtering features. The encoder maps the second filtering features to a low-dimensional latent space representation, and the decoder reconstructs the features while minimizing the reconstruction error. This optimizes and reassembles the features, ultimately obtaining core data features of dimension k, where k... m, the corresponding core data feature is .Should It is used as input for subsequent scene classification models to achieve efficient and accurate determination of the current driving scene.

[0040] Based on the above technical solution, this application embodiment performs a hierarchical processing flow of feature extraction, chi-square filtering, variance filtering, and noise reduction reconstruction sequentially on the multimodal raw input data. This extracts core data features that can accurately characterize the essential attributes of driving scenarios. Based on these core data features, the current driving scenario is determined, which effectively improves the accuracy and reliability of current driving scenario recognition. This provides an accurate and solid basis for the adaptive selection of subsequent recommendation strategies, avoids recommendation strategy matching errors caused by scenario recognition bias, and ensures the effectiveness and stability of the entire service location recommendation process.

[0041] S130: Determine whether the current driving scenario belongs to the preset driving scenario set, and obtain the determination result.

[0042] The preset driving scenario set refers to a known set of scenarios established by the in-vehicle system through historical learning or manual predefinition. The scenarios in this set typically include, but are not limited to, typical scenarios such as regular commuting, long-distance travel, short-distance urban travel, driving with low battery / low fuel, driver fatigue, driving in inclement weather, and driving near service areas / commercial areas.

[0043] Specifically, if the current driving scenario belongs to the preset set of driving scenarios, it means that the current driving scenario is a typical scenario known to the system and has sufficient historical user interaction data; if the current driving scenario does not belong to the preset set of driving scenarios, it means that the scenario is a new scenario not covered by the system or a cold start scenario, lacking sufficient historical behavioral data support. This judgment result directly determines which recommendation strategy to adopt to adapt to the cognitive state of the current scenario.

[0044] S140: Based on the judgment result, determine the target recommendation strategy from the first recommendation strategy and the second recommendation strategy. The first recommendation strategy is a recommendation strategy based on reinforcement learning algorithm, and the second recommendation strategy is a recommendation strategy based on large language model reasoning.

[0045] The target recommendation strategy is a specific strategy ultimately selected based on the judgment result and used for the current service recommendation decision. The target recommendation strategy is one of the first recommendation strategy and the second recommendation strategy, that is, it is either the first recommendation strategy or the second recommendation strategy.

[0046] Specifically, the first recommendation strategy is built upon a reinforcement learning algorithm. Leveraging the algorithm's closed-loop iterative learning mechanism across driving scenarios, service location recommendations, and user experience feedback, a complete decision-making training space is constructed. Through continuous iterative optimization of the driving scenario-service location recommendation algorithm, the potential patterns of user preferences in different scenarios are fully explored, achieving accurate capture and dynamic optimization of long-term user preferences. This provides the ability for continuous iterative upgrades of recommendation decisions driven by data, and can output highly relevant personalized recommendation results in the current scenario. The second recommendation strategy is built upon the semantic understanding and reasoning capabilities of a large language model. This strategy relies on the model's massive knowledge base and complex semantic parsing capabilities. After inputting the current driving scenario into the large language model, the model analyzes the core needs and potential preferences of the scenario, thereby generating a service recommendation scheme adapted to the user in the current scenario. Based on the above judgment results, the in-vehicle system selects one recommendation strategy from the first and second recommendation strategies as the target recommendation strategy to achieve accurate service recommendations in different scenarios.

[0047] In one possible implementation, determining the target recommendation strategy from the first and second recommendation strategies based on the judgment result includes the following steps: If the judgment result indicates that the current driving scenario belongs to the preset driving scenario set, the first recommendation strategy will be determined as the target recommendation strategy. If the judgment result indicates that the current driving scenario does not belong to the preset driving scenario set, the second recommendation strategy is determined as the target recommendation strategy.

[0048] Among them, the preset driving scenario set refers to a set of typical and high-frequency driving scenarios that are pre-summarized, defined and stored.

[0049] Specifically, after obtaining the judgment result, the in-vehicle system executes strategy selection logic to determine the target recommendation strategy. When the judgment result indicates that the current driving scenario belongs to the preset driving scenario set, that is, the current driving scenario is identified as a known driving scenario with sufficient historical interaction data, the first recommendation strategy is determined as the target recommendation strategy, thereby activating the execution process of the first recommendation strategy and loading the reinforcement learning model parameters and service recommendation rules corresponding to the matched preset scenario. When the judgment result indicates that the current driving scenario does not belong to the preset driving scenario set, that is, the current driving scenario is identified as a newly emerging or unknown driving scenario with scarce historical data, the second recommendation strategy is determined as the target recommendation strategy, thereby activating the execution process of the second recommendation strategy, calling the large language model inference engine and preparing the structured prompt word generation module. Thus, the in-vehicle system dynamically determines the target recommendation strategy adopted for this service recommendation.

[0050] Based on the above technical solution, this application embodiment dynamically selects different recommendation methods according to the judgment result: in known driving scenarios, a recommendation method based on reinforcement learning is adopted; in unknown driving scenarios, a reasoning method based on a large language model is switched, realizing dynamic adaptation of recommendation strategies and scenario states. This ensures the accuracy and personalization of recommendation decisions in known driving scenarios and improves the effectiveness of recommendation services in unknown driving scenarios, providing a reliable basis for subsequent recommendation of service locations and effectively improving the accuracy of service location recommendations and user experience.

[0051] In one possible implementation, in some implementations of the first aspect, when the target recommendation strategy is a first recommendation strategy, the above-mentioned recommendation of target service locations to users based on the target recommendation strategy includes the following steps: The set of services to be recommended is determined based on the current driving scenario. The set of services to be recommended includes multiple candidate service locations. Obtain preset scores for multiple candidate service locations. These preset scores are the user's past experience ratings for the candidate service locations. The current driving scenario, multiple candidate service locations, and the preset scores corresponding to the multiple candidate service locations are respectively used as the scene space state, immediate reward, and recommendation action in the reinforcement learning algorithm to obtain the recommendation scores corresponding to each of the multiple candidate service locations. The target service location is determined based on the recommendation scores of each of the multiple candidate service locations and the user's past preference for multiple candidate service locations.

[0052] The set of services to be recommended refers to a collection of service locations that are considered to have potential recommendation value and meet basic screening criteria (such as being within a preset distance and being operational) in the current driving scenario. The preset score is a quantitative score derived from analyzing users' historical behavior data, representing the user's satisfaction with past experiences at each candidate service location. This score integrates explicit user feedback (such as ratings and reviews) and implicit behaviors (such as usage frequency and dwell time), and serves as an estimate of the immediate reward that the corresponding recommendation action (i.e., recommending the location) can obtain in the reinforcement learning algorithm. The target service location refers to the service location that will ultimately be proactively recommended to the user.

[0053] Specifically, when the target recommendation strategy is the first recommendation strategy, the process begins by determining the set of services to be recommended based on the current driving scenario. First, a pre-defined scenario-service mapping database is retrieved. This database is built based on historical service data and user demand analysis, and pre-defines service types suitable for different typical driving scenarios (e.g., breakfast shops, temporary parking lots, fast charging stations, etc. for commuting scenarios). Then, combined with the points of interest information along the route from the environmental context data, all service locations within a pre-defined range (e.g., within 5 kilometers) of the current vehicle location, in operation, and conforming to the service type of that scenario are selected, forming a set of services to be recommended containing multiple candidate service locations. Subsequently, when obtaining the pre-defined scores corresponding to multiple candidate service locations, user experience feedback data from previous use of each candidate service location is extracted from the user behavior database. This includes explicit ratings (e.g., star ratings of 1-5 points) and implicit feedback (e.g., usage duration, number of repeated selections). These are then weighted and converted into a unified, quantified pre-defined score (range 0-10 points).

[0054] Based on this, the above elements are mapped into the framework of a reinforcement learning algorithm: the current driving scenario is taken as state s in the state space, each candidate service location in the set of services to be recommended is taken as action a in the action space, and the corresponding preset score is directly used as the immediate reward r obtained after performing the action. The Q-learning algorithm is used to construct the state-action (sa) value function. After initializing the Q-table, the Q-value is iteratively updated according to the following formula:

[0055] Where s represents the current state, a represents the action chosen in the current state, α is the learning rate, γ is the discount factor, and r is the immediate reward. Let Q be the largest Q value among all possible actions a' in the next state s'.

[0056] After a predetermined number of iterations of training, the recommended score for each candidate service location is obtained, which is its corresponding Q value. Finally, based on the recommended scores (and their respective Q values) of the multiple candidate service locations, and combined with the user's past preferences for various types of service locations, the in-vehicle system determines the target service location from the multiple candidate service locations.

[0057] Based on the above technical solution, this application integrates driving scenarios, candidate service locations, and historical experience scores into a reinforcement learning algorithm to calculate the recommendation score for each candidate service location. Then, it combines this with the user's past preferences for each candidate service location to determine the target service location. This allows the algorithm to automatically learn service preferences from the user's actual historical interactions without relying on preset rules. This approach fully utilizes the user's past usage preferences and service feedback, while incorporating user feedback into the recommendation score calculation process. This makes service recommendations in known driving scenarios more aligned with the user's personalized needs, effectively improving the alignment between the recommendation results and the user's actual needs, thereby ensuring the accuracy of service location recommendations.

[0058] In one possible implementation, in some implementations of the first aspect, determining the target service location based on the recommendation scores corresponding to each of the multiple candidate service locations and the user's past preference for the multiple candidate service locations includes the following steps: The candidate service location with the highest score is selected from the recommended scores of multiple candidate service locations, and the first service location is obtained. If the preference level of the first service location is greater than the preset preference level threshold, the first service location will be determined as the target service location.

[0059] The first service location refers to the single candidate service location with the highest recommendation score selected from multiple candidate service locations after ranking them according to the recommendation scores calculated by the reinforcement learning algorithm. The preset preference threshold is a preference quantification threshold (such as 0.7, corresponding to 70% preference weight) that the system pre-sets based on users' historical preference data, industry service recommendation benchmarks, and scenario adaptation requirements. It is used to determine whether the first service location matches users' long-term usage habits.

[0060] Specifically, when determining the first service location from multiple candidate service locations, the vehicle system first extracts the recommendation scores calculated by the reinforcement learning algorithm for each candidate service location, sorts them in descending order of score, and selects the candidate service location with the highest score as the first service location. If there are multiple candidate service locations with the same recommendation score and the highest score, the real-time service quality parameters of each location (such as real-time user evaluation, current queuing time, and remaining resource status) are further compared, and the candidate service location with the best real-time service quality is selected as the first service location.

[0061] When determining whether the preference level for the first service location exceeds a preset preference threshold, the personal scenario priority matrix is ​​first retrieved. This matrix is ​​dynamically updated using an information gain algorithm, with the update formula as follows:

[0062] in, It is the information entropy of scene s. It is the conditional entropy of scene s given feature A.

[0063] This matrix records the user's preference weights for various service locations in the current driving scenario (weight values ​​range from 0 to 1, dynamically updated by the user's historical selection records and interaction feedback data using an information gain algorithm). The preference weight corresponding to the first service location in the matrix is ​​extracted as its preference level. This preference level is then compared with a preset preference level threshold. If the preference level is greater than the preset preference level threshold, it means that the first service location not only meets the optimal recommendation solution for the current scenario but also matches the user's long-term preferences, and it is directly determined as the target service location. If the preference level is less than or equal to the preset preference level threshold, the next candidate service location is selected from the remaining candidate service locations in descending order of recommendation score. The above preference level verification steps are repeated until a service location with a preference level greater than the preset threshold is selected as the target service location. If the preference level of all candidate service locations does not exceed the preset threshold, the first service location with the highest recommendation score is determined as the target service location, and a preference adjustment prompt is pushed to the user simultaneously, inviting the user to provide feedback on whether they accept the recommendation. Subsequently, the personal scenario priority matrix and the preset preference level threshold are updated based on user feedback.

[0064] Based on the above technical solution, this application embodiment employs a two-stage screening and judgment process to ensure that the service location recommendation results are suitable for the current driving scenario and align with users' long-term usage habits. First, the highest-scoring service location from the candidate service location recommendation scores output by the reinforcement learning algorithm is selected as the first service location, ensuring the recommendation result matches the current context. Then, the user's historical preference level for the first service location is compared with a preset preference threshold. Only when the preference level of the first service location exceeds this threshold is it determined as the target service location. This effectively avoids the problem of relying solely on algorithmic scoring for recommendations while ignoring actual user habits. The recommendation results not only meet scenario adaptation standards but also closely reflect users' real needs and preferences, thereby significantly improving the accuracy of service location recommendations in known driving scenarios and optimizing the user's actual experience.

[0065] In one possible implementation, in some implementations of the first aspect, when the target recommendation strategy is a second recommendation strategy, the above-mentioned recommendation of target service locations to users based on the target recommendation strategy includes the following steps: Transform the current driving scenario into structured text prompts; The text prompt information is input into the large language model, and the large language model outputs the user's requirements. From multiple preset service locations, a second service location is obtained by determining the preset service location that meets the user's needs. The target service location is determined based on the second service location.

[0066] The second service location refers to a set of one or more preset service locations that, after being selected from multiple preset service locations, meet the user needs of the large language model's inference output.

[0067] Specifically, when converting the current driving scenario into structured text prompts, the core features of the current driving scenario (including key vehicle parameters, user behavior tendencies, and key environmental context information) are first extracted and then structured according to a preset prompt template. The template includes three parts: "scenario type + core constraints + potential demand orientation". For example, "Scenario: New user driving in an unfamiliar area; Vehicle condition: Pure electric vehicle with 25% remaining battery power; Environment: Currently experiencing heavy rain, driving on a suburban road; Demand orientation: Matching service locations that meet the needs of energy replenishment, rain shelter, and temporary rest", ensuring that the text prompts are logically clear, with no key information omitted, and adapting to the semantic understanding requirements of a large language model. After the text prompts are input into the large language model, the model, based on its built-in knowledge base and semantic reasoning capabilities, analyzes the core constraints and potential needs of the current scenario and outputs a structured description of user needs (such as "needs to provide fast charging services for electric vehicles, have an indoor rest area, support temporary parking, and have a service location within 10 kilometers of the current location"). If the needs output by the model are ambiguous or ambiguous, the in-vehicle system will guide the model to reason a second time by supplementing prompts (such as "focusing only on the most urgent core needs in the current scenario") until it outputs clear and actionable user needs.

[0068] When determining a second service location from multiple preset service locations, the vehicle system first retrieves preset service location data (including service type, function configuration, location information, real-time status, etc.) synchronized with the local service resource library and third-party service interfaces. Then, it breaks down the user requirements output by the large language model into quantifiable filtering conditions (e.g., service type includes "electric vehicle fast charging," function configuration includes "indoor rest area," distance less than or equal to 10 kilometers, current operating status is "open"). Through multi-condition matching, all preset service locations meeting the requirements are filtered to form the second service location. If no service location meets the criteria after filtering, the vehicle system adjusts the filtering threshold (e.g., widening the distance threshold to 15 kilometers) or guides the large language model to re-optimize the user requirements (e.g., "prioritize matching locations with fast charging services, indoor rest areas are optional"), and executes the filtering process again to obtain a second service location. Finally, based on the obtained second service locations, a target service location is determined. From the second service locations, the service location with the best fit for the current driving scenario and user requirements is selected and designated as the target service location.

[0069] Based on the above technical solution, this application embodiment, when facing unknown or newly emerging driving scenarios, no longer relies on user historical data. Instead, it converts the current situation into structured text prompts and inputs them into a large language model. The large language model understands and infers the services the user might need, and then filters candidate service locations that meet the requirements from preset service locations. This ensures that reasonable recommendations can be given even without historical interaction records, avoiding the problems of traditional recommendation methods that "cannot recommend" or "recommend randomly" in new users, new routes, or unexpected situations. This effectively improves the accuracy of service location recommendations and user experience in new scenarios.

[0070] In one possible implementation, in some implementations of the first aspect, determining the target service location based on the second service location includes the following steps: If the second service location is a single location, then the second service location will be designated as the target service location. If there are multiple second service locations, then calculate the evaluation score for each second service location; The second service location with the highest recommended score from all the recommended scores is selected as the third service location; The third service location was identified as the target service location.

[0071] The third service location refers to the second service location with the highest evaluation score when there are multiple second service locations. This third service location is selected by calculating and sorting the recommended scores of each second service location.

[0072] Specifically, when there is only one second service location, that second service location is directly determined as the optimal solution matching the user's needs, without any additional calculations, and is directly identified as the target service location.

[0073] When there are multiple secondary service locations, a dynamic weighting mechanism is used to calculate the evaluation score for each location. The calculation formula is as follows:

[0074] in It is the weight of the i-th scene attribute. It is an indicator function (value 1 when the service matches attribute i, otherwise 0), and β is the context weight coefficient. The weights are calculated from the environmental context.

[0075] The evaluation score for each second service location is calculated based on this formula (i.e. After calculation, all second service locations are sorted from highest to lowest according to their evaluation scores, and the location ranked first is determined as the third service location. If there are ties for the highest score, a second comparison is made by referring to real-time information related to the environmental context (such as the availability of resources at the service location and the travel time of vehicles to the location) to finally determine the third service location and use it as the target service location.

[0076] Based on the above technical solution, this application embodiment adopts a flexible filtering method for the candidate service locations output by the large language model under the second recommendation strategy. When a single service location is matched, it is directly adopted to ensure recommendation efficiency. When there are multiple candidate service locations that meet the user's needs, the optimal option is selected by calculating the evaluation score. It does not randomly select from multiple candidate service locations, thereby avoiding random or unsatisfactory recommendation results in the case of multiple candidates. This makes the service location recommendation in the new scenario more reliable and further improves the accuracy of the recommendation.

[0077] S150: Recommend target service locations to users based on target recommendation strategies.

[0078] Specifically, when the target recommendation strategy is the first recommendation strategy, the recommendation is executed based on a reinforcement learning algorithm: First, the corresponding set of services to be recommended is retrieved from the local service database according to the current driving scenario. This set contains multiple candidate service locations. Then, a preset score is obtained for each candidate service location. The preset score is generated based on the user's historical feedback behavior on similar services (such as clicks, visits, ratings). Next, the current driving scenario is taken as the state, each candidate service location is taken as the action, and the preset score is taken as the immediate reward. These are input into a trained reinforcement learning model (such as a Q-learning network) to calculate the recommendation value of each candidate service location and obtain their respective recommendation scores. Finally, the candidate service locations are comprehensively ranked by combining the recommendation score with the user's long-term preference weight in this scenario (such as the value from the personal scenario priority matrix), and the service location with the highest ranking is selected as the target service location for push.

[0079] When the target recommendation strategy is the second recommendation strategy, the recommendation is executed based on the large language model reasoning: First, the core data features of the current driving scenario are transformed into structured text prompts, which contain key elements of vehicle status, user behavior tendencies, and environmental context. Then, the prompts are input into the large language model, which outputs a description of the user's potential service needs based on its semantic understanding and common sense reasoning capabilities (such as "needs a charging station and a service area with a rest area"). Next, the description of needs is parsed, and preset service locations that meet the conditions are matched to obtain one or more candidate service locations. If there is only one matching result, it is directly determined as the target service location. If there are multiple matching results, the evaluation score of each candidate location is further calculated, and the one with the highest score is selected as the target service location for push notification.

[0080] Based on the above technical solution, this application embodiment achieves accurate judgment of the current driving scenario by collecting multimodal data of vehicle condition, user behavior and environmental context, and selects different recommendation strategies according to the judgment results. This avoids the problem that traditional methods rely on only a single recommendation strategy and are difficult to effectively recommend service locations when facing unknown driving scenarios. It ensures that reliable service location recommendations can be provided in both known and unknown driving scenarios, thereby significantly improving the accuracy of service location recommendations and user experience.

[0081] In one possible implementation, after determining the target service location, the in-vehicle system will also construct a unified service call specification based on the Interface Definition Language (IDL), automatically converting the recommendation results into data formats adapted to the interface requirements of different service providers, thus achieving standardization and compatibility of cross-platform service calls. Furthermore, the in-vehicle system will dynamically decide the timing and interaction format of the recommended content output based on preset push activation conditions, balancing driving safety and user experience. These push activation conditions include the real-time distance between the vehicle and the target service location, the user's current driving state (e.g., whether they are fatigued, distracted, or operating under heavy load), whether navigation guidance is activated, and the complexity of the current road environment. When fatigued driving or prolonged continuous driving is detected, the system will prioritize proactively pushing the nearest service location with rest facilities via voice broadcast, avoiding visual interaction methods such as pop-ups and flashing prompts that may interfere with driving. When the user is in a stable state, driving at a low speed, or waiting in a parked position, a lightweight interface combining text and graphics can be used to ensure effective information delivery while minimizing interference with driving tasks.

[0082] Furthermore, a multimodal feedback evaluation mechanism is employed to quantitatively analyze the effectiveness of each recommendation. This method integrates user sentiment evaluation indicators and behavioral evaluation indicators. Sentiment evaluation indicators include user satisfaction, interest, and trust in the recommendation results, which can be inferred through in-vehicle voice tone analysis, facial expression recognition, or explicit voice evaluation (such as "This recommendation is good"). Behavioral evaluation indicators include user engagement with the recommended content (such as whether they view details), acceptance (such as whether they click to confirm or initiate navigation), execution accuracy (such as whether they actually go to the service location), and rejection behavior (such as immediately closing the prompt, explicitly rejecting via voice, or remaining unresponsive for an extended period). The in-vehicle system structures these multi-source feedback signals, distinguishing between positive adoption samples and negative rejection samples, and marking them as valid recommendation data and erroneous recommendation data, respectively. Subsequently, this labeled data is fed back into the training process of the reinforcement learning model as a supervisory signal for policy gradient updates, continuously optimizing the state-action value function. This allows the recommendation strategy to adaptively learn the evolution of user preferences and scene response patterns, thereby continuously improving accuracy, timeliness, and user acceptance in subsequent service recommendations.

[0083] Figure 2 This illustration shows a schematic diagram of the architecture of a service location recommendation method provided in an embodiment of this application, such as... Figure 2 As shown, Figure 2The service location recommendation architecture showcased comprises four modules: First, a data acquisition module collects vehicle condition, user behavior, and environmental context data as the basic input for recommendations. Next, a scenario analysis module processes and integrates the collected data, extracting core features to determine the specific scenario. Then, a recommendation decision module provides two parallel recommendation paths based on the determined scenario: one path generates recommendations using reinforcement learning and a priority matrix, while the other uses a large language model and dynamic weight stacking. Finally, a service execution module handles the implementation of external service integration, recommended content delivery, and service execution status monitoring based on the recommendations output from the above paths. This comprehensive architecture realizes a complete recommendation logic from data input to scenario recognition, multi-path decision-making, and service implementation.

[0084] Figure 3 This application provides a schematic diagram of the structure of a service location recommendation device according to an embodiment of the present application. Figure 3 As shown, the service venue recommendation device 300 includes: Data acquisition module 310 is used to acquire raw input data, which includes vehicle condition data, user behavior data and environmental context data; The scene determination module 320 is used to determine the current driving scene based on the original input data; The scene judgment module 330 is used to determine whether the current driving scene belongs to the preset driving scene set and obtain the judgment result. The strategy determination module 340 is used to determine the target recommendation strategy from the first recommendation strategy and the second recommendation strategy based on the judgment result. The first recommendation strategy is a recommendation strategy based on the reinforcement learning algorithm, and the second recommendation strategy is a recommendation strategy based on the reasoning of the large language model. The service recommendation module 350 is used to recommend target service locations to users based on the target recommendation strategy.

[0085] In one possible implementation, the scene determination module 320 is used for: Feature extraction is performed on the original input data to obtain the original data features; The original data features are subjected to chi-square filtering to obtain the first filtered features; The first filtering feature is subjected to variance filtering to obtain the second filtering feature; The second filtered feature is subjected to noise reduction and data reconstruction to obtain the core data feature; The current driving scenario is determined based on core data features.

[0086] In one possible implementation, the strategy determination module 340 is used for: If the judgment result indicates that the current driving scenario belongs to the preset driving scenario set, the first recommendation strategy will be determined as the target recommendation strategy. If the judgment result indicates that the current driving scenario does not belong to the preset driving scenario set, the second recommendation strategy is determined as the target recommendation strategy.

[0087] In one possible implementation, the service recommendation module 350 is used for: The set of services to be recommended is determined based on the current driving scenario. The set of services to be recommended includes multiple candidate service locations. Obtain preset scores for multiple candidate service locations. These preset scores are the user's past experience ratings for the candidate service locations. The current driving scenario, multiple candidate service locations, and the preset scores corresponding to the multiple candidate service locations are respectively used as the scene space state, immediate reward, and recommendation action in the reinforcement learning algorithm to obtain the recommendation scores corresponding to each of the multiple candidate service locations. The target service location is determined based on the recommendation scores of each of the multiple candidate service locations and the user's past preference for multiple candidate service locations.

[0088] In one possible implementation, the service recommendation module 350 is used for: The candidate service location with the highest score is selected from the recommended scores of multiple candidate service locations, and the first service location is obtained. If the preference level of the first service location is greater than the preset preference level threshold, the first service location will be determined as the target service location.

[0089] In one possible implementation, the service recommendation module 350 is used for: Transform the current driving scenario into structured text prompts; The text prompt information is input into the large language model, and the large language model outputs the user's requirements. From multiple preset service locations, a second service location is obtained by determining the preset service location that meets the user's needs. The target service location is determined based on the second service location.

[0090] In one possible implementation, the service recommendation module 350 is used for: If the second service location is a single location, then the second service location will be designated as the target service location. If there are multiple second service locations, then calculate the evaluation score for each second service location; The second service location with the highest recommended score from all the recommended scores is selected as the third service location; The third service location was identified as the target service location.

[0091] It should be noted that the service location recommendation device provided in the above embodiments is only illustrated by the division of the above functional modules when executing the service location recommendation method. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the service location recommendation device and the service location recommendation method embodiments provided in the above embodiments belong to the same concept. Therefore, for details not disclosed in the device embodiments of this application, please refer to the embodiments of the service location recommendation method of this application, which will not be repeated here.

[0092] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0093] Figure 4 This application provides a schematic diagram of the structure of a vehicle according to an embodiment of the present application. Figure 4 As shown, the vehicle 400 includes a memory 401 and a processor 402, wherein the memory 401 stores executable program code 4011, and the processor 402 is used to call and execute the executable program code 4011, which is a service location recommendation method.

[0094] This embodiment can divide the vehicle into functional modules according to the above method example. For example, each function can be assigned to a separate module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0095] The vehicle provided in this embodiment is used to execute the service location recommendation method described above, and therefore can achieve the same effect as the above implementation method.

[0096] The vehicle may include a processing module and a storage module. The processing module is used to control and manage the vehicle's actions. The storage module is used to support the vehicle in executing relevant program code and data.

[0097] The processing module may be a processor or a controller, which can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. The processor may also be a combination of functions that implement computing capabilities, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and a microprocessor, etc., and the storage module may be a memory.

[0098] In addition, the vehicle provided in the embodiments of this application may specifically be a chip, component or module. The vehicle may include a connected processor and a memory. The memory is used to store instructions. When the vehicle is running, the processor may call and execute the instructions to make the chip execute a service location recommendation method in the above embodiments.

[0099] This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to implement a service location recommendation method in the above embodiment.

[0100] The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, Digital Video Discs (DVDs), Compact Disc Read-Only Memory (CD-ROMs), microdrives, and magneto-optical disks, read-only memory (ROMs), random access memory (RAMs), erasable programmable read-only memory (EPROMs), electrically erasable programmable read-only memory (EEPROMs), dynamic random access memory (DRAMs), video random access memory (VRAMs), flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0101] This embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned steps to implement the service location recommendation method provided in the above embodiment.

[0102] In this embodiment, the vehicle, computer-readable storage medium, computer program product, or chip are all used to execute the corresponding service location recommendation method provided above. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding service location recommendation method provided above, and will not be repeated here.

[0103] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0104] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0105] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for recommending service locations, characterized in that, The method includes: Acquire raw input data, which includes vehicle condition data, user behavior data, and environmental context data; The current driving scenario is determined based on the original input data; Determine whether the current driving scenario belongs to a preset set of driving scenarios, and obtain the determination result; Based on the judgment result, a target recommendation strategy is determined from the first recommendation strategy and the second recommendation strategy. The first recommendation strategy is a recommendation strategy based on a reinforcement learning algorithm, and the second recommendation strategy is a recommendation strategy based on large language model reasoning. Based on the target recommendation strategy, target service locations are recommended to users.

2. The method according to claim 1, characterized in that, Determining the current driving scenario based on the original input data includes: Feature extraction is performed on the original input data to obtain the original data features; The original data features are subjected to chi-square filtering to obtain the first filtered features; The first filtering feature is subjected to variance filtering to obtain the second filtering feature; The second filtered feature is subjected to noise reduction and data reconstruction to obtain the core data feature; The current driving scenario is determined based on the core data features.

3. The method according to claim 1, characterized in that, The step of determining the target recommendation strategy from the first recommendation strategy and the second recommendation strategy based on the judgment result includes: If the determination result indicates that the current driving scenario belongs to the preset driving scenario set, the first recommendation strategy is determined as the target recommendation strategy; If the judgment result indicates that the current driving scenario does not belong to the preset driving scenario set, the second recommendation strategy is determined as the target recommendation strategy.

4. The method according to claim 1, characterized in that, When the target recommendation strategy is the first recommendation strategy, recommending target service locations to the user based on the target recommendation strategy includes: A set of services to be recommended is determined based on the current driving scenario, and the set of services to be recommended includes multiple candidate service locations; Obtain preset scores corresponding to the multiple candidate service locations, where the preset scores are the user's past experience ratings for the candidate service locations; The current driving scenario, the multiple candidate service locations, and the preset scores corresponding to the multiple candidate service locations are respectively used as the scene space state, immediate reward, and recommendation action in the reinforcement learning algorithm to obtain the recommendation scores corresponding to each of the multiple candidate service locations. The target service location is determined based on the recommendation scores of each of the multiple candidate service locations and the user's past preference for the multiple candidate service locations.

5. The method according to claim 4, characterized in that, The step of determining the target service location based on the recommendation scores corresponding to each of the multiple candidate service locations and the user's past preference for the multiple candidate service locations includes: The candidate service location with the highest score is determined from the recommended scores of each of the multiple candidate service locations, and the first service location is obtained. If the preference level of the first service location is greater than the preset preference level threshold, the first service location is determined as the target service location.

6. The method according to claim 1, characterized in that, When the target recommendation strategy is the second recommendation strategy, recommending target service locations to the user based on the target recommendation strategy includes: The current driving scenario is converted into structured text prompts. The text prompt information is input into the large language model, and the large language model outputs the user's requirements. A second service location is obtained by determining a pre-set service location that meets the user's needs from multiple pre-set service locations; The target service location is determined based on the second service location.

7. The method according to claim 6, characterized in that, Determining the target service location based on the second service location includes: If the second service location is a single location, then the second service location is identified as the target service location; If there are multiple second service locations, then calculate the evaluation score for each second service location; The second service location with the highest recommended score from among the recommended scores of each of the second service locations is determined to obtain the third service location; The third service location is identified as the target service location.

8. A service venue recommendation device, characterized in that, The service location recommendation device includes: The data acquisition module is used to acquire raw input data, which includes vehicle condition data, user behavior data, and environmental context data. The scene determination module is used to determine the current driving scene based on the original input data; The scene determination module is used to determine whether the current driving scene belongs to a preset driving scene set and obtain the determination result; The strategy determination module is used to determine a target recommendation strategy from the first recommendation strategy and the second recommendation strategy based on the judgment result. The first recommendation strategy is a recommendation strategy based on a reinforcement learning algorithm, and the second recommendation strategy is a recommendation strategy based on large language model reasoning. The service recommendation module is used to recommend target service locations to users based on the target recommendation strategy.

9. A vehicle, characterized in that, The vehicles include: Memory, used to store executable program code; A processor is configured to call and run the executable program code from the memory, causing the vehicle to perform the service location recommendation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed, implements the service location recommendation method as described in any one of claims 1 to 7.

Images