Vehicle-mounted music recommendation method, device and equipment and readable storage medium

Abstract

The invention relates to the technical field of vehicles, and discloses a vehicle-mounted music recommendation method, device and equipment and a readable storage medium, and the method comprises the steps: carrying out the preprocessing of vehicle-mounted environment data, user related data and vehicle-mounted operation data, and generating feature information; constructing a user emotion evolution track based on the feature information, and generating feature fusion information in combination with a preset public emotion map; inputting the feature fusion information into a music recommendation model to generate a candidate music set; and updating the candidate music set according to the user real-time feedback information to obtain target recommended music. According to the method provided by the invention, the individuation and real-time performance of vehicle-mounted music recommendation can be improved, the adaptability with a driving scene is enhanced, and the user experience and the driving comfort are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a method, apparatus, device and readable storage medium for recommending music in vehicles. Background Technology

[0002] With the increasing integration of intelligent driving and in-vehicle entertainment, music recommendation, as a crucial component of in-vehicle human-machine interaction, has become a key means of enhancing the driving experience and alleviating driving stress. Existing music recommendation systems largely rely on users' historical listening records or platform rating data for modeling. While these systems can reflect users' music preferences to some extent in general application scenarios, they suffer from insufficient adaptability in driving scenarios. Theoretically, the driving process is not only influenced by an individual's emotional state but also dynamically correlated with environmental factors such as vehicle speed, weather, and road conditions, as well as physiological signals such as heart rate variability and skin conductance. Without comprehensive perception and processing of multimodal information, recommendation results often fail to accurately match the actual needs of users in different driving situations.

[0003] In detail, existing music recommendation technologies still have several shortcomings: First, the data dimension is too narrow, relying solely on users' historical behavior data and neglecting the combined effects of environmental information and physiological signals in driving scenarios, resulting in poor contextual adaptability; second, the emotion modeling method is crude, often using fixed discrete labels (such as "pleasant / sad"), failing to capture the gradual changes in users' emotions during driving; third, the recommendation mechanism is lagging, lacking real-time feedback channels and unable to adjust recommendation strategies based on users' immediate interactions or physiological reactions; fourth, there is a lack of visualization mechanisms, failing to intuitively demonstrate the correspondence between music consumption and emotional changes. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, apparatus, device, and readable storage medium for recommending in-vehicle music, which can effectively solve the problem that existing in-vehicle music recommendations lack comprehensive perception of environment, emotion, and operating status, resulting in insufficient personalization and scene adaptability of recommendation results.

[0005] In a first aspect, embodiments of this application provide a method for recommending in-vehicle music, including: Acquire in-vehicle environmental data, user-related data, and in-vehicle operation data; The in-vehicle environment data, the user-related data, and the in-vehicle operation data are preprocessed to generate feature information; Based on the aforementioned feature information, a user's emotional evolution trajectory is constructed, and feature fusion information is generated by combining it with a preset mass emotion map. The fused feature information is input into the music recommendation model to generate a candidate music set; The candidate music set is updated based on real-time user feedback to obtain the target recommended music.

[0006] In some embodiments, acquiring in-vehicle environment data, user-related data, and in-vehicle operation data includes: Vehicle-mounted sensors collect air and weather information inside and outside the vehicle to generate vehicle environmental data. By collecting playback behavior, switching behavior, and favorite behavior through user terminals, and combining this with information collected by wearable devices, user-related data is formed. Vehicle operating status parameters are collected through the vehicle control interface to form vehicle operating data.

[0007] In some embodiments, the preprocessing of the in-vehicle environment data, the user-related data, and the in-vehicle operation data to generate feature information includes: Outlier detection and denoising processing are performed on the in-vehicle environment data, the user-related data, and the in-vehicle operation data to obtain multimodal data; The multimodal data is aligned according to the time dimension to obtain a synchronized data sequence; Feature extraction is performed based on the synchronized data sequence to generate feature information representing user emotions and environmental states.

[0008] In some embodiments, constructing a user's emotional evolution trajectory based on the feature information and generating feature fusion information by combining it with a preset mass emotion graph includes: The feature information is input into a temporal modeling network to generate a user emotion evolution trajectory that changes over time. By associating the user's emotional evolution trajectory with the public sentiment graph, the association information between individual user and group emotions is obtained. Feature fusion is performed based on the associated representation information to generate feature fusion information.

[0009] In some embodiments, inputting the feature fusion information into a music recommendation model to generate a candidate music set includes: The feature fusion information is input into a preset music recommendation model to obtain an initial set of candidate music. The initial set of candidate music is filtered according to emotional matching degree and scene adaptability to obtain the filtered candidate music set.

[0010] In some embodiments, updating the candidate music set based on real-time user feedback to obtain target recommended music includes: Collect real-time feedback information from users, compare the real-time feedback information with the matching results of the candidate music set, and generate correction instructions; The candidate music set is updated according to the correction instructions to obtain the target recommended music.

[0011] In some embodiments, the method further includes: Record the user-related data and in-vehicle environment data corresponding to the target recommended music to form a historical data set; The music recommendation model is optimized and trained based on the historical data set.

[0012] Secondly, embodiments of this application provide an in-vehicle music recommendation device, comprising: The data acquisition module is used to acquire in-vehicle environmental data, user-related data, and in-vehicle operation data. The data processing module is used to preprocess the vehicle environment data, the user-related data, and the vehicle operation data to generate feature information; The trajectory processing module is used to construct the user's emotional evolution trajectory based on the feature information, and generate feature fusion information by combining it with a preset mass emotion map. The music recommendation module is used to input the feature fusion information into the music recommendation model to generate a candidate music set; The update module is used to update the candidate music set based on real-time user feedback to obtain the target recommended music.

[0013] Thirdly, embodiments of this application provide an electronic device, the electronic device including a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the in-vehicle music recommendation method of the first aspect described above.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium, wherein when the computer program is executed on a processor, it implements the in-vehicle music recommendation method of the first aspect described above.

[0015] The embodiments of this application have the following beneficial effects: They acquire in-vehicle environment data, user-related data, and in-vehicle operation data, and perform outlier detection, noise reduction, and time alignment processing to generate feature information representing user emotions and environmental states; based on the feature information, they construct the user's emotional evolution trajectory and combine it with a preset public sentiment graph to generate feature fusion information; they input the feature fusion information into a music recommendation model to generate a candidate music set, and filter it based on emotion matching degree and scene adaptability; further, they dynamically update the candidate music set based on user feedback to obtain the target recommended music; simultaneously, they record the user-related data and in-vehicle environment data corresponding to the target recommended music to form a historical data set, and optimize and train the recommendation model. Through this solution, accurate modeling of user emotional states and the external environment can be achieved in in-vehicle scenarios, and the multi-dimensional matching ability of recommendations can be improved by combining a public sentiment graph, thereby dynamically adjusting the recommendation results to make the output target recommended music more suitable for the user's real-time preferences and driving environment. This not only improves the personalization and accuracy of music recommendations but also has the ability to iteratively optimize with historical data, thus effectively improving the in-vehicle music recommendation experience. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart of an embodiment of the in-vehicle music recommendation method of this application is shown; Figure 2 This paper illustrates another flowchart of the in-vehicle music recommendation method according to an embodiment of this application; Figure 3 This paper illustrates another flowchart of the in-vehicle music recommendation method according to an embodiment of this application; Figure 4 This diagram illustrates a structural schematic of an in-vehicle music recommendation method according to an embodiment of this application. Detailed Implementation

[0018] The technical solutions in 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, and not all embodiments.

[0019] The components of the embodiments of this application described and illustrated in the accompanying drawings can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] In the following text, the terms "comprising," "having," and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more combinations thereof. Furthermore, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0021] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0022] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0023] Given the shortcomings of existing in-vehicle music recommendation systems in comprehensively modeling the environment, user emotions, and vehicle operating status, resulting in poor performance in terms of real-time performance and personalization, this application proposes an in-vehicle music recommendation method. This method involves multi-source data collection and preprocessing, emotional trajectory construction and graph fusion, candidate music screening and feedback updates, and model optimization training using historical data to achieve dynamic adjustment and accurate matching of music recommendation results in in-vehicle scenarios.

[0024] The following examples illustrate the in-vehicle music recommendation method.

[0025] Figure 1 A flowchart of an in-vehicle music recommendation method according to an embodiment of this application is shown. Exemplarily, the in-vehicle music recommendation method includes the following steps: Step S100: Obtain vehicle environment data, user-related data, and vehicle operation data.

[0026] Among them, vehicle environment data refers to raw data representing the state of the internal and external environment of the vehicle acquired by vehicle-mounted sensors; user-related data refers to data representing user interaction behavior and physiological signals collected by user terminals; and vehicle operation data refers to parameters reflecting the vehicle's operating conditions acquired through the vehicle interface. In this embodiment, environmental data acquisition, user data acquisition, and operation data acquisition are initiated simultaneously in the vehicle scenario to facilitate data preprocessing in subsequent steps. For example, the ability to represent road types and geographical environments can be enhanced by acquiring fused positioning information from GPS, Wi-Fi, and cellular base stations and synchronizing it with environmental data.

[0027] In an optional embodiment, step S100 includes the following sub-steps: S101 collects air and weather information inside and outside the vehicle through vehicle-mounted sensors to form vehicle environmental data.

[0028] The vehicle-mounted sensing devices include temperature and humidity sensors, barometers, and light intensity sensors to characterize air quality and weather conditions. Exemplarily, they can collect air temperature, humidity, PM2.5 concentration, air pressure, and external light levels to form environmental data entries. For example, in rainy weather, the vehicle-mounted sensing devices can detect changes in humidity and light intensity, storing these parameters as vehicle environmental data for feature extraction in subsequent steps. Rainfall levels or precipitation intensity indices (0–100) from weather stations can also be obtained via an interface to quantify the impact of weather on driving scenarios.

[0029] S102 collects playback behavior, switching behavior, and favorite behavior through the user terminal, and combines them with information collected by wearable devices to form user-related data.

[0030] In this context, the user terminal refers to a mobile device running a music application that records the user's interactive behavior; wearable devices refer to smartwatches or bracelets with sensing capabilities, used to collect the user's physiological signals. For example, the user terminal can record tracks played for more than 30 seconds as valid playback behavior, track skipping for less than 5 seconds as negative interaction feedback, and record the user's favorites and sharing actions within the application. In other words, the in-vehicle system establishes a connection with the user's wearable device via Bluetooth or wireless communication. After the connection is established, it collects heart rate variability and skin conductance response signals, and combines them with the interactive behavior data to form user-related data for unified preprocessing in subsequent steps.

[0031] In other implementations, the user terminal also records user-created playlists and comment interactions, using these as auxiliary features of long-term user preferences; and the in-vehicle terminal can be configured to automatically establish a Bluetooth connection when a paired device is detected nearby, thereby reducing the latency of physiological data access. For example, user authorization and device binding are required when connecting to a wearable device for the first time; when authorization fails or the connection is interrupted, the system switches to a user-related data collection mode based solely on interaction behavior.

[0032] S103 collects vehicle operating status parameters through the vehicle control interface to form vehicle operating data.

[0033] The vehicle control interface refers to a vehicle bus interface conforming to the OBD-II standard, used to transmit operating parameters of vehicle sensors. For example, it can collect real-time vehicle speed, engine speed, steering wheel grip force signals, and wiper status, and convert them into structured operational data. For instance, when the vehicle is traveling at high speed and the wipers are on, the system can simultaneously generate operational data for both vehicle speed and wiper status, which can then be preprocessed along with environmental and user-related data in subsequent steps. In other implementations, parameters such as turn signal activation, brake pedal opening, and acceleration can be further acquired to refine the scenario adaptability assessment.

[0034] Step S200: Preprocess the vehicle environment data, user-related data, and vehicle operation data to generate feature information.

[0035] Preprocessing refers to the process of improving the quality and unifying the encoding of multi-source heterogeneous raw data, including outlier detection and denoising, temporal alignment, and feature extraction. For example, outlier identification and noise suppression are performed on multimodal data, followed by alignment of data streams with different sampling rates on the same time axis. Finally, a set of feature vectors representing emotion and environmental state is extracted to facilitate the construction and fusion modeling of emotion trajectories in subsequent steps. For instance, outliers can be removed first using an anomaly detection strategy, and temporal alignment can employ flexible registration methods to eliminate rate differences before outputting structured feature tensors for the model. Furthermore, for acquired external text data sources (such as music platform comments), distributed crawling and cleaning can be performed first, and sentence-level sentiment intensity (0–1) can be calculated using SnowNLP as auxiliary labels for the sentiment side.

[0036] In an optional embodiment, step S200 includes the following sub-steps: S201 performs outlier detection and noise reduction on in-vehicle environmental data, user-related data, and in-vehicle operation data to obtain multimodal data.

[0037] Outlier detection refers to identifying and removing data samples that significantly deviate from the normal distribution; denoising refers to suppressing random disturbances introduced by motion, sensor fluctuations, or communication jitter. Exemplarily, the Isolation Forest algorithm is used to perform outlier detection on in-vehicle environmental data, user-related data, and in-vehicle operation data, and Kalman filtering is applied to suppress noise in time-series signals affected by motion artifacts, in order to obtain multimodal data of controllable quality.

[0038] S202 aligns the multimodal data according to the time dimension to obtain a synchronized data sequence.

[0039] Temporal alignment refers to registering data sequences from different sources and with different sampling rates under a unified time index, ensuring that data items at the same time correspond to each other. As an example, Dynamic Time Warping (DTW) is used to register multi-source sequences under local elastic matching constraints, outputting synchronized data sequences with a unified timestamp for feature extraction in subsequent steps. In other implementations, for environmental parameter sequences and physiological signal sequences with different sampling frequencies, the optimal matching path can be calculated using DTW first, and then resampled to a unified time axis based on this path, preserving the relative order of key inflection points.

[0040] S203 extracts features based on synchronized data sequences to generate feature information that represents user emotions and environmental states.

[0041] Feature extraction refers to mapping the synchronized original sequence to a low-dimensional or structured description space to characterize the target emotion and scene attributes. For example, a five-dimensional emotion vector e=[V,A,D,T,E] (Valence, Arousal, Dominance, Tension, Energy) can be constructed: V (Valence): Valence, reflecting the "positive or negative valence" of the emotion (e.g., happiness is "positive valence," sadness is "negative valence"); A (Arousal): Arousal level, which reflects the "level of excitement" of an emotion (e.g., excitement is "high arousal", and calmness is "low arousal"); D (Dominance): Sense of control, reflecting the degree of "control" over emotions or situations (e.g., "high sense of control" when confident, "low sense of control" when helpless); T (Tension): Tension level, reflecting the "degree of tension" of emotions (e.g., "high tension" when anxious, "low tension" when relaxed); E (Energy): The sense of energy, which reflects the "energy level" in one's emotions (for example, being full of energy is "high energy", and being tired is "low energy").

[0042] The environmental elements are quantified and encoded (for example, weather conditions are mapped to a “rain index” and precipitation intensity is quantified in the range of 0–100) to generate a set of feature information for modeling, so as to model and fuse the trajectory of emotion evolution in subsequent steps.

[0043] Step S300: Construct the user's emotional evolution trajectory based on feature information, and generate feature fusion information by combining it with a preset mass emotion map.

[0044] The user emotion evolution trajectory is a temporal representation describing how user emotions change over time. The mass sentiment graph is a dynamic graph structure built based on the similarity relationship of emotion vectors, with users and songs as nodes, and updated within a time window. Feature fusion information refers to the feature set after jointly encoding the user-side temporal representation and the mass sentiment graph representation. Exemplarily, the preprocessed feature information is temporally modeled to obtain the emotion evolution trajectory, which is then associated with a pre-defined mass sentiment graph to obtain an association representation. Based on this, feature fusion information is generated so that it can be input into the recommendation model in subsequent steps. For example, the mass sentiment graph can adopt a dynamic graph structure, with nodes containing users and songs, edge weights calculated from the cosine similarity of emotion vectors, and retaining data from the past 30 days using a sliding window strategy.

[0045] In one alternative embodiment, such as Figure 2 As shown, step S300 includes the following sub-steps: S301 inputs feature information into a time-series modeling network to generate a trajectory of user emotion evolution over time.

[0046] Temporal modeling networks refer to neural network structures capable of representing time-dependent relationships, used to extract time-evolving emotion representations from synchronous data sequences. Exemplarily, feature information can be input sequentially into a bidirectional long short-term memory network or a Transformer-XL network or Bi-LSTM network based on long-range dependency processing capabilities, outputting an emotion evolution trajectory arranged by time index, which can then be associated with a public sentiment graph in subsequent steps. For example, the emotion vector can be represented in five dimensions [V, A, D, T, E] corresponding to Valence, Arousal, Dominance, Tension, and Energy. The trajectory output by the network can describe the temporal change from "depression – stability – relief," or express state transition probabilities in the form of a Markov chain, used for subsequent graph association and fusion.

[0047] S302, link the user's emotional evolution trajectory with the public emotion graph to obtain the correlation representation information of individual user emotions and group emotions.

[0048] Here, the association representation information refers to the joint representation obtained by aligning and aggregating user-side temporal emotions with group-side node / edge relationships under graph structure constraints. Exemplarily, on a graph with users and songs as nodes, edge weights are set based on the similarity relationship of emotion vectors, and attention-weighted message aggregation is performed on the set of adjacent nodes of the user's emotion evolution trajectory within the corresponding time window for feature fusion in subsequent steps. For example, the graph neural network can use GraphSAGE and introduce an attention mechanism to achieve aggregation, in the form of:

[0049] in, Represents a node In the Layer features (updated feature representation); Activation functions (such as Sigmoid and ReLU) introduce nonlinear transformations to enhance the model's expressive power and attention coefficients; This represents summing the results over the neighboring nodes u of node v. yes (the set of neighbors); This represents the attention weight of node u to node v (reflecting the importance of neighbor u to v, which is adaptively learned by the attention mechanism). represents the learnable weight matrix of the l-th layer, used to perform a linear transformation on the features of neighbor node u, mapping them to a new dimension; The features of node u at layer l (original / intermediate features before update).

[0050] S303, feature fusion is performed based on association representation information to generate feature fusion information.

[0051] Feature fusion refers to the weighted combination or distillation mapping of two types of representations to obtain a unified feature space representation that can be used for recommendation reasoning, while maintaining the consistency between individual user temporal features and group graph features. Exemplarily, the attention module of the application scenario assigns dynamic weights to group information under different scenarios, and combines teacher-student knowledge distillation constraints to generate feature fusion information, which is then input into the trained music recommendation model in subsequent steps to generate a candidate set. For example, the group side can act as a teacher model to impose soft objective constraints on the individual-side temporal network, while contrast constraints are introduced during the distillation process to maintain the scene similarity structure, thereby obtaining fused features that take into account both individual preferences and group knowledge transfer.

[0052] Step S400: Input the feature fusion information into the music recommendation model to generate a candidate music set.

[0053] Feature fusion information refers to the unified features obtained by jointly encoding user-side temporal sentiment representation and mass sentiment graph representation, used to characterize the comprehensive relationship between individual preferences and group knowledge. The music recommendation model is a learning model that scores input features and outputs candidate results. Exemplarily, after obtaining the feature fusion information, it is input into the trained recommendation model for inference, resulting in a candidate list with relevance scores, and forming a candidate music set for subsequent filtering. For example, the recommendation model can introduce graph distillation and scene attention during the training phase to improve generalization ability, and combine reinforcement learning strategies (such as PPO) for inference control during the online phase, thereby obtaining candidate results that take into account both individual and group semantics.

[0054] In an optional embodiment, step S400 includes the following sub-steps: S401, Input the feature fusion information into the preset music recommendation model to obtain the initial set of candidate music.

[0055] The preset music recommendation model refers to an inference model trained offline and deployed in an in-vehicle or cloud environment. This model calculates the relevance of the input feature fusion information and outputs initial candidate results. The initial set of candidate music refers to the original output set of the model before being filtered by scene or rule constraints. Exemplarily, the feature fusion information is fed into the model, and an initial candidate set is generated based on the model's internal relevance score or reward estimate. The corresponding scores and necessary information are retained for subsequent filtering according to predetermined indicators. For example, during the training phase, the model can receive teacher signals from a public sentiment graph for knowledge distillation, and during inference, it can simultaneously use individual temporal sentiment trajectories and graph neighborhood aggregation features to calculate relevance scores, thereby generating the initial candidate set.

[0056] S402, the initial set of candidate music is filtered according to the degree of emotional matching and scene adaptability to obtain the filtered set of candidate music.

[0057] Here, emotion matching degree refers to the degree of matching between the emotion tag or emotion vector of the candidate track and the user's current emotion characteristics; scene adaptability refers to the degree of adaptability between the attributes of the candidate track and the current driving environment or scene template. Exemplarily, emotion matching degree is calculated based on the user's current emotion vector and the emotion tag of the candidate track, and scene adaptability is calculated based on environment coding or a scene template library. The initial set is then filtered using these two indicators to obtain a filtered set of candidate music, which can be updated in subsequent steps based on real-time user feedback. For example, the filtering rules may include: an emotion tag matching degree greater than 80% is a mandatory condition; a tempo BPM in the range of 60–80 is a preferred condition; and tracks containing "intense" sound effects are excluded. Scene adaptability can be determined by referring to typical scene templates generated by the DBSCAN algorithm. In a specific example, under high-speed or complex road conditions, an upper limit threshold can be set for tempo intensity, peak loudness, or frequency band energy proportion; tracks exceeding the threshold will not enter the candidate set.

[0058] Step S500: Update the candidate music set based on real-time user feedback to obtain the target recommended music.

[0059] Real-time user feedback refers to user-side information generated instantly during the music recommendation process, including both implicit and explicit feedback. Implicit feedback refers to behavioral signals exhibited by users in natural interactions, such as song skipping, playback duration, and volume adjustment, reflecting the user's potential acceptance of the recommendation results. Explicit feedback refers to adjustment information actively provided by users through the interactive interface, such as dragging the five-dimensional emotion slider or selecting emotion tags, directly representing the user's emotional needs. For example, after the candidate music set is generated, the system monitors users' song skipping behavior (skipping less than 5 seconds is considered negative feedback) and playback duration (playback exceeding 30 seconds is considered positive feedback) in real time. Simultaneously, it receives adjustment information input by users through the five-dimensional emotion slider. This feedback is compared with the emotion matching degree of the candidate music to form update conditions, allowing for the correction of the candidate set and the output of target recommended music in subsequent steps. For instance, when a user adjusts the Valence value to 0.7 using the slider, the ranking weight of "positive" tracks can be increased accordingly, while tracks that are quickly switched can be de-weighted or removed to improve the matching degree between the recommendation results and the user's immediate preferences.

[0060] In other implementations, a time-axis-based emotion change curve can be displayed on the playback control interface and smoothed using spline interpolation to visually present changes in emotion indicators before and after recommendations; user emotion trajectories can also be projected in a three-dimensional Valence–Arousal–Tension space, or scene heatmaps based on geographic location and time period can be generated to display the distribution of track preferences.

[0061] In an optional embodiment, step S500 includes the following sub-steps: S501 collects real-time feedback information from users, compares the real-time feedback information with the matching results of the candidate music set, and generates correction instructions.

[0062] The matching result refers to the calculated correlation between candidate music and the current user's emotional and environmental characteristics; the correction instruction refers to control information that adjusts the weights, rankings, or filtering conditions of the candidate set based on real-time feedback. For example, the user's feedback at the current moment is quantified as an adjustment parameter, compared with the matching degree of the candidate music's emotional label or emotional vector, forming an increase / decrease instruction for the track category or label weight, so that the candidate set can be updated in subsequent steps. For instance, when the user adjusts Valence to +0.7, the emotional matching degree of the candidate tracks is immediately recalculated, and an increase coefficient (e.g., increasing the weight by 30%) is applied to tracks with the "happy" label, thus correcting the original recommendation "SadSong" to "Happy Now," while recording the comparison data before and after the adjustment.

[0063] S502, update the candidate music set according to the correction instruction to obtain the target recommended music.

[0064] Here, "updating" refers to adjusting the members and ranking of the candidate set according to correction instructions, including increasing the weight of tracks that meet the feedback instructions, decreasing the weight of tracks that do not meet the conditions, or performing elimination operations. The target recommended music refers to the music that meets the current matching requirements after the update and is used for playback output. Exemplarily, the candidate set is rearranged and screened according to the correction instructions, the target recommended music is output, and key quantities during the update process (such as weight changes, matching degree changes, and feedback parameters) are stored for data recording and strategy optimization in subsequent steps. For example, after completing the target recommendation, the feedback and the changes in physiological data before and after can be written into historical data and used for updating online reinforcement learning strategies (such as PPO) or periodically distilled into the user-side model to improve the stability and response speed of recommendations in subsequent scenarios.

[0065] In another alternative embodiment, such as Figure 3 As shown, this in-car music recommendation method also includes the following steps: Step S600: Record the user-related data and in-vehicle environment data corresponding to the target recommended music to form a historical data set.

[0066] The historical dataset refers to the multimodal information collected and stored during the user's playback of the target recommended music, reflecting the correspondence between the recommendation results and the user's state and environmental conditions. User-related data refers to data information related to user behavior and physiological state. In-vehicle environmental data refers to external and internal environmental parameters acquired by in-vehicle sensors. For example, when a user plays the target recommended music, the user's playback duration, switching behavior, and physiological signals such as heart rate variability and skin conductance are synchronously recorded along with environmental parameters such as vehicle speed, weather conditions, and air humidity, generating corresponding historical data entries for use in model optimization in subsequent steps. For instance, if a user plays "Happy Now" while driving at high speed in rainy weather and exhibits a stable heart rate, this scenario, along with the corresponding physiological parameters and interaction data, is stored as historical data for subsequent training.

[0067] Step S700: Optimize and train the music recommendation model based on the historical data set to improve the matching effect of the target recommended music.

[0068] The optimization training refers to the process of updating the parameters and improving the performance of the existing recommendation model using historical data sets, ensuring a higher degree of fit between the recommendation results and the user's emotional state and environmental state. The music recommendation model refers to a model that combines individual user trajectories and public graph information for candidate generation and ranking. Demonstratively, the accumulated historical data set is periodically input into the recommendation model, and the model parameters are jointly optimized through supervision signals and a contrastive loss function. This enables the model to more accurately predict users' emotional preferences in subsequent inferences, thus continuously outputting highly matched recommendation results in later steps. For example, a dynamic weight allocation strategy (e.g., adjusting classification / contrast loss weights according to the accuracy of the validation set) can be used during the optimization process, and the LAMB optimizer (learning rate 3e-5, 1000-step warm-up, weight decay 0.01) can be used to complete the parameter updates. Furthermore, the emotional intensity of external comments can be used as an auxiliary supervision signal in training, optimized together with the main task in the form of a multi-task loss.

[0069] Figure 4 A schematic diagram of a vehicle music recommendation device according to an embodiment of this application is shown. Exemplarily, the vehicle music recommendation device 100 includes: Data acquisition module 110 is used to acquire in-vehicle environment data, user-related data and in-vehicle operation data; Data processing module 120 is used to preprocess the vehicle environment data, the user-related data and the vehicle operation data to generate feature information; The emotion trajectory processing module 130 is used to construct the user's emotion evolution trajectory based on the feature information, and generate feature fusion information by combining it with a preset mass emotion map. Music recommendation module 140 is used to input the feature fusion information into the music recommendation model to generate a candidate music set; The update module 150 is used to update the candidate music set based on real-time user feedback information to obtain the target recommended music.

[0070] It is understood that the apparatus of this embodiment corresponds to the method of the above embodiments, and the options in the above embodiments are also applicable to this embodiment, so they will not be described again here.

[0071] This application also provides an electronic device, exemplary in that the electronic device includes a processor and a memory, wherein the memory stores a computer program, and the processor, by running the computer program, causes the electronic device to perform the functions of the various modules in the above-described method or apparatus.

[0072] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0073] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving execution instructions.

[0074] This application also provides a computer-readable storage medium for storing the computer program used in the aforementioned electronic device. For example, the computer-readable storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0075] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0076] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0077] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0078] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes 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.

Claims

1. A method for recommending music in a car, characterized in that, The method includes: Acquire in-vehicle environmental data, user-related data, and in-vehicle operation data; The in-vehicle environment data, the user-related data, and the in-vehicle operation data are preprocessed to generate feature information; Based on the aforementioned feature information, a user's emotional evolution trajectory is constructed, and feature fusion information is generated by combining it with a preset mass emotion map. The fused feature information is input into the music recommendation model to generate a candidate music set; The candidate music set is updated based on real-time user feedback to obtain the target recommended music.

2. The in-vehicle music recommendation method according to claim 1, characterized in that, The acquisition of in-vehicle environment data, user-related data, and in-vehicle operation data includes: Vehicle-mounted sensors collect air and weather information inside and outside the vehicle to generate vehicle environmental data. By collecting playback behavior, switching behavior, and favorite behavior through user terminals, and combining this with information collected by wearable devices, user-related data is formed. Vehicle operating status parameters are collected through the vehicle control interface to form vehicle operating data.

3. The in-vehicle music recommendation method according to claim 1, characterized in that, The preprocessing of the in-vehicle environment data, the user-related data, and the in-vehicle operation data to generate feature information includes: Outlier detection and denoising processing are performed on the in-vehicle environment data, the user-related data, and the in-vehicle operation data to obtain multimodal data; The multimodal data is aligned according to the time dimension to obtain a synchronized data sequence; Feature extraction is performed based on the synchronized data sequence to generate feature information representing user emotions and environmental states.

4. The in-vehicle music recommendation method according to claim 1, characterized in that, The step of constructing a user's emotional evolution trajectory based on the feature information and generating feature fusion information by combining it with a preset mass emotion map includes: The feature information is input into a temporal modeling network to generate a user emotion evolution trajectory that changes over time. By associating the user's emotional evolution trajectory with the public sentiment graph, the association information between individual user and group emotions is obtained. Feature fusion is performed based on the associated representation information to generate feature fusion information.

5. The in-vehicle music recommendation method according to claim 1, characterized in that, The step of inputting the feature fusion information into the music recommendation model to generate a candidate music set includes: The feature fusion information is input into a preset music recommendation model to obtain an initial set of candidate music. The initial set of candidate music is filtered according to emotional matching degree and scene adaptability to obtain the filtered candidate music set.

6. The in-vehicle music recommendation method according to claim 1, characterized in that, The step of updating the candidate music set based on real-time user feedback to obtain the target recommended music includes: Collect real-time feedback information from users, compare the real-time feedback information with the matching results of the candidate music set, and generate correction instructions; The candidate music set is updated according to the correction instructions to obtain the target recommended music.

7. The in-vehicle music recommendation method according to claim 1, characterized in that, The method further includes: Record the user-related data and in-vehicle environment data corresponding to the target recommended music to form a historical data set; The music recommendation model is optimized and trained based on the historical data set.

8. A car music recommendation device, characterized in that, include: The data acquisition module is used to acquire in-vehicle environmental data, user-related data, and in-vehicle operation data. The data processing module is used to preprocess the vehicle environment data, the user-related data, and the vehicle operation data to generate feature information; The trajectory processing module is used to construct the user's emotional evolution trajectory based on the feature information, and generate feature fusion information by combining it with a preset mass emotion map. The music recommendation module is used to input the feature fusion information into the music recommendation model to generate a candidate music set; The update module is used to update the candidate music set based on real-time user feedback to obtain the target recommended music.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the in-vehicle music recommendation method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed on a processor, implements the in-vehicle music recommendation method according to any one of claims 1-7.

Images