Machine learning based in-vehicle information delivery decision system and method

Abstract

The application discloses a vehicle-mounted information putting decision system and method based on machine learning, and belongs to the technical field of machine learning and vehicle-mounted information processing. In order to solve the technical problem that the prior art cannot output adaptive putting strategy according to real-time situation, and the accuracy of putting opportunity, priority adjustment and conflict prediction is insufficient, the application discloses a vehicle-mounted information putting decision system and method based on machine learning, content feature extraction is performed on to-be-put content, situation feature coding is performed on vehicle environment data, and system resource features are combined for fusion reasoning, so that putting priority, putting opportunity and strategy parameters are output, adaptive putting decision of vehicle-mounted information is realized, and the ability of vehicle-mounted multimedia processing intercut conflict is improved.

Description

Technical Field

[0001] This invention relates to the fields of machine learning and vehicle information processing technology, and in particular to a machine learning-based vehicle information delivery decision system and method. Background Technology

[0002] In order to upgrade traditional public transportation spaces into information-based mobile service platforms, and to create new commercial value and social benefits while improving the quality of public services, AI (Artificial Intelligence) technology is combined with traditional public transportation in-vehicle multimedia. Intelligent information delivery also conforms to the development trend of smart cities and transportation modernization.

[0003] Currently, AI-driven bus information delivery decision-making systems generally consist of "three layers and two cores". The three layers include the perception layer, the cognition layer, and the decision and execution layer, while the two cores include the cloud data center and the scheduling and advertising platform. The core idea of ​​this architecture is that AI identifies the current bus operation and riding situation, then automatically matches the most suitable information content, and then intelligently decides whether and how to insert the message.

[0004] The perception layer typically deploys various vehicle sensors, such as GPS (Global Positioning System), gyroscopes, cameras, microphones, and OBD (On-Board Diagnostics), to collect various environmental data, such as weather, road conditions, and passenger density. The cognition layer uses vehicle context recognition AI to identify location, events, environment, and passenger status, and uses content semantic understanding AI to identify the type, semantics, and theme of the playback material. Then, when encountering a broadcast interruption, the decision-making and execution layer uses decision-making AI to determine whether to interrupt, the content to be interrupted, and the timing of the interruption.

[0005] Among them, the vehicle context recognition AI is used to identify bus operation status, geographical location, time, weather, passenger flow, road conditions, etc.; the content semantic understanding AI performs semantic classification and topic tag extraction for the content to be broadcast; the delivery decision AI integrates context, content features and business rules to output the final insertion decision and priority; in addition, it includes passenger feature recognition AI to analyze passenger gender ratio, age group and emotional state in order to adjust the content; and scene matching AI to match the current context with the semantic vector of the content library to calculate the optimal delivery item.

[0006] When combining various AI technologies, the typical implementation process of current mainstream public transportation AI multimedia systems is as follows: First, data is collected. Then, AI based on context recognition identifies the current location, weather, passenger flow, and operating status. Next, semantic analysis of the content library is performed to extract tags for each piece of content (such as "safety tips," "commercial advertisements," and "guide audio"). Then, AI based on scene matching calculates the content that best matches the current context based on semantic similarity. Finally, AI for decision-making evaluates priorities, taking into account factors such as safety priority, passenger experience, and advertising revenue. Finally, the playback scheduling system executes the insertion of content to control the timing, duration, conflict detection, and playback feedback, thereby achieving dynamic and precise content delivery.

[0007] The above-mentioned technology has at least the following technical problems:

[0008] Existing in-vehicle information delivery solutions lack a joint learning and fusion decision-making mechanism that considers the characteristics of the content to be delivered, the vehicle context, and the system resources. This makes it difficult to output adaptive delivery strategies based on real-time contexts, resulting in insufficient accuracy in delivery timing, priority adjustment, and conflict prediction. Summary of the Invention

[0009] To address the technical problem of existing technologies that struggle to output adaptive delivery strategies based on real-time contexts, leading to insufficient accuracy in delivery timing, priority adjustment, and conflict prediction, this invention extracts features of the content to be delivered using a first trained model, predicts delivery conflict feasibility using a second trained model, and extracts vehicle context features using a third trained model. It then fuses and infers the content features, context features, and system resource features to output delivery priority, delivery timing, and delivery strategy parameters. The technical solution provided in the embodiment is as follows: On the one hand, a machine learning-based in-vehicle information delivery decision-making system is provided, including: a content feature extraction module, used to predict delivery conflicts of the insertion signal if an insertion signal is received when the in-vehicle multimedia is playing information according to the current delivery task queue order, in order to determine the feasibility of insertion; a context feature encoding module, used to perform context recognition and context delivery correlation evaluation based on AI when performing delivery conflict prediction, so as to determine whether to adjust the insertion data before the delivery conflict prediction; and a strategy output module, used to restore the system according to the task type of the currently playing information after the insertion ends.

[0010] On the other hand, a machine learning-based in-vehicle information delivery decision-making method is provided. The method includes: when the in-vehicle multimedia is playing information according to the current delivery task queue order, if an insertion signal is received, the insertion signal is predicted to determine the feasibility of insertion; when predicting the delivery conflict, AI is used to perform context recognition and context delivery correlation assessment to determine whether to adjust the insertion data before the delivery conflict prediction; after the insertion ends, the information is restored according to the task type of the currently playing information.

[0011] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. The in-vehicle information delivery decision system based on machine learning provided by this invention, when the in-vehicle multimedia system is playing information according to the current delivery task queue order, if an insertion signal is received, it performs delivery conflict prediction on the insertion signal to determine the feasibility of insertion. This makes up for the shortcomings of traditional in-vehicle multimedia systems, where insertion requests often directly interrupt the current playback task, causing abrupt stops in the picture or voice, audio misalignment, etc., thereby helping to ensure the accuracy of in-vehicle information delivery decisions, and making the delivery of in-vehicle multimedia information smoother, more fluid and intelligent.

[0012] 2. When predicting delivery conflicts, AI is used for context recognition and context-based delivery relevance assessment to determine whether to adjust the insertion data before the conflict prediction. This fills the gap in traditional conflict analysis, which only relies on the delivery task queue and time and lacks environmental awareness, making it impossible to determine whether the inserted content is relevant to the current scene. This improves the system's information relevance, context adaptability, and user experience quality. Furthermore, after the insertion ends, the system is restored according to the task type of the current playback information, improving the system's intelligent task management level and realizing a complete closed loop in the in-vehicle multimedia information playback process. This effectively solves the problems that often occur in existing traditional systems after the insertion ends, such as the inability to accurately restore playback tasks, chaotic delivery task queues, or content loss.

[0013] 3. This invention obtains an analytical value for quantifying the impact of insertion on system operation by performing feasibility calculations on the acquired data. This value is recorded as the delivery conflict prediction coefficient. This solves the problem that traditional systems cannot quantitatively measure the impact of insertion operations on system performance, providing a unified quantitative basis for subsequent delay control and insertion timing decisions. This enables refined control of the system's insertion behavior. The coefficient is then compared with an insertion conflict threshold: if the delivery conflict prediction coefficient is greater than the threshold, an insertion conflict occurs. The deviation between the delivery conflict prediction coefficient and the threshold is mapped to obtain the insertion delay duration. After the delay duration is reached, the insertion algorithm is activated to insert information, preventing the inability of existing technologies to flexibly adjust insertion timing according to the severity of the conflict, which leads to unstable playback or excessive information delay. This not only avoids system stuttering or playback errors caused by forced insertion under high load but also implements a system-level "insertion buffer protection mechanism," enhancing system stability. If the delivery conflict prediction coefficient is not greater than the insertion conflict threshold, the insertion algorithm is immediately activated, thus complementing the delay mechanism and achieving intelligent dynamic scheduling.

[0014] 4. By acquiring the context relevance index of AI-powered insertion information and weighting and fusing it with preset insertion context weights to obtain the corresponding insertion context relevance evaluation value, the system fills the gap in the existing system's inability to achieve differentiated insertion priority control under different vehicles and operating strategies, enabling flexible insertion decision-making strategy configuration. Then, the insertion context relevance evaluation value is compared with the context relevance threshold, solving the problem that traditional systems cannot determine whether the matching degree between the inserted content and the context meets the triggering conditions and lack quantitative definition standards. This realizes a context-adaptive insertion triggering mechanism, making the system's delivery more intelligent and accurate. If the insertion context relevance evaluation value is higher than the context relevance threshold, the insertion information's delivery priority is adjusted; otherwise, the delivery priority of the insertion information is not adjusted. This compensates for the shortcomings of traditional systems, where the insertion priority is fixed and cannot be dynamically adjusted according to the actual environment, resulting in delays or low-priority playback of important content. In this way, the overall intelligence of the in-vehicle multimedia system is improved. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A schematic diagram of the structure of a machine learning-based vehicle information delivery decision system provided in an embodiment of this application; Figure 2 A schematic diagram illustrating the delivery conflict prediction process provided in an embodiment of this application; Figure 3 A comparison diagram illustrating the optimization of delivery conflict prediction provided in an embodiment of this application; Figure 4 A schematic diagram illustrating the process of evaluating the relevance of vehicle contextual insertion information provided in this application embodiment; Figure 5 This is a flowchart illustrating the method of a machine learning-based vehicle information delivery decision system provided in an embodiment of this application. Detailed Implementation

[0017] The following provides explanations for some of the terms used in this application. It should be noted that these explanations are for the convenience of those skilled in the art and do not constitute a limitation on the scope of protection claimed in this application.

[0018] The embodiments of this application involve at least one, including one or more; where "multiple" means two or more. Furthermore, it should be understood that in the description of this specification, terms such as "first," "second," and "third" are used only for descriptive purposes and should not be construed as indicating relative importance or order. For example, "first device" and "second device" do not represent the degree of importance of the two or their order, but are merely for descriptive distinction. In the embodiments of this application, "and / or" merely describes an association relationship, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0019] The directional terms mentioned in the embodiments of this application, such as "up", "down", "left", "right", "inner", and "outer", are only for reference to the directions in the accompanying drawings. Therefore, the directional terms used are for better and clearer explanation and understanding of the embodiments of this application, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this application.

[0020] References to "one embodiment," "in some examples," or "some embodiments" as described in the embodiments of this application mean that one or more embodiments of this specification include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in some examples," "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0021] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0022] Example 1 like Figure 1 The diagram shown is a structural schematic of a machine learning-based vehicle information delivery decision-making system provided in an embodiment of this application. The system includes: The content feature extraction module is used to identify the type of multimedia data through the first AI recognition model and generate corresponding delivery priority data. When the in-vehicle multimedia plays information according to the current delivery task queue order, if an insertion signal is received, the module predicts the delivery conflict of the insertion signal to determine the feasibility of insertion, thereby ensuring the accuracy of in-vehicle information delivery decisions.

[0023] The context feature encoding module is used to predict delivery conflicts through the second AI decision model. When predicting delivery conflicts, the third AI context model is used to identify the current vehicle environment data. Based on AI, context identification and context delivery correlation assessment are performed to determine whether to adjust the insertion data before delivery conflict prediction.

[0024] The strategy output module is used to restore playback after the interruption ends, based on the task type of the current playback information. Task types include, but are not limited to, short interruptions, high-priority advertisements, emergency broadcasts, and multicast task overlays.

[0025] By setting up a content feature extraction module, a context feature encoding module, and a strategy output module, the intelligent scheduling and smooth control of the in-vehicle information delivery decision-making system are realized. Its beneficial effects are as follows: First, the content feature extraction module, by monitoring the system occupancy rate, current playback progress, and historical insertion conflict rate in real time, assesses the feasibility of insertion requests, helping to avoid output interruptions and information loss caused by system resource overload or playback task conflicts, thus ensuring the continuity and stability of the in-vehicle multimedia playback process. Second, the context feature encoding module introduces an AI context recognition and semantic association evaluation mechanism, which can dynamically adjust the insertion strategy based on the vehicle's environment, driving status, time, and the semantic matching degree of the content, realizing intelligent and contextualized information delivery and improving the relevance and dissemination value of the inserted content. Finally, the strategy output module automatically restores the playback status according to the task type, supporting seamless connection of multiple tasks such as short-term insertions, high-priority advertisements, and emergency broadcasts, ensuring the logicality of the playback order and the continuity of the user experience. This invention effectively improves the intelligent decision-making capability, playback smoothness, and content accuracy of the in-vehicle information delivery decision-making system.

[0026] The first AI recognition model is used to identify the type of multimedia information corresponding to the inserted signal and generate an initial delivery priority. This model is built on a deep convolutional neural network or a visual transformer (ViT) to handle video / image inserted content; for text / speech inserted content, it is built on a pre-trained language model (such as BERT or RoBERTa). The training process is as follows: First, historical inserted information samples are collected and manually labeled with type tags, such as "emergency broadcast," "commercial advertisement," "public service advertisement," and "guide information," while assigning an initial priority value to each type according to business rules. Then, the sample data is input into the pre-trained model for fine-tuning, using a cross-entropy loss function to optimize the model parameters so that it can accurately identify the type of inserted information. After training, the model can classify newly received inserted information in real time and output the corresponding delivery priority.

[0027] Through the above methods, the first AI recognition model has achieved automated understanding and initial priority judgment of inserted content, providing a data foundation for subsequent prediction and dynamic adjustment of delivery conflicts.

[0028] It should be further explained that the second AI decision model for insertion feasibility analysis and the third AI context model for context recognition involved in this application are not simple logical judgment units, but computational models pre-trained by machine learning algorithms based on historical operating data.

[0029] Specifically, the construction process of the second AI decision-making model is as follows: First, the system operation logs for the historical time period are collected, and system resource feature data, including CPU utilization, GPU decoding load, memory utilization, and network bandwidth utilization, are extracted, along with the corresponding insertion result labels, i.e., whether a conflict occurred or not. Second, the above feature data and labels are used as a training set and input into an initial model based on gradient boosting decision trees or deep neural networks for supervised training. During the training process, cross-entropy is used as the loss function, and the model parameters are iteratively updated through the backpropagation algorithm until the model's accuracy on the validation set reaches a preset threshold. Finally, a conflict prediction weight matrix is ​​generated for real-time calculation of the conflict prediction coefficient.

[0030] Similarly, the third AI context model is based on the Transformer architecture, such as the BERT model, and is pre-trained and fine-tuned on a large-scale vehicle broadcast corpus and environmental sensor data, so that it can accurately understand the correlation between semantics such as accidents ahead and slippery roads in rainy weather and the current GPS location and weather sensor data, and output a standardized correlation score.

[0031] like Figure 2 The diagram shows a flowchart of the delivery conflict prediction process provided in this application embodiment. The specific logic is as follows: First, obtain the insertion conflict assessment data, then perform insertion feasibility calculation based on the insertion conflict assessment data to obtain the delivery conflict prediction coefficient, and compare it with the insertion conflict definition threshold: If the delivery conflict prediction coefficient is greater than the insertion conflict definition threshold, it indicates that an insertion conflict has occurred. Based on the deviation between the delivery conflict prediction coefficient and the insertion conflict definition threshold, a mapping is performed to obtain the insertion delay duration, and the insertion algorithm is activated to insert the insertion information after the insertion delay duration is reached; if the delivery conflict prediction coefficient is not greater than the insertion conflict definition threshold, the insertion algorithm is activated immediately to insert the information. Through the above process, not only can playback stuttering, black screen, or task interruption problems caused by resource overload be avoided, but the stability of the information release process is also guaranteed.

[0032] It should be noted that the process of predicting insertion conflicts for the inserted signal to determine the feasibility of insertion is as follows: First, acquire interstitial conflict assessment data, which includes system usage data, current playback progress, and historical interstitial conflict rates. System usage data includes one or more of the following: CPU utilization, GPU decoding load, network bandwidth utilization, and memory utilization. When the system load is concentrated on a single resource type, or when the interstitial content is of a single type and only single system usage data is used (e.g., if the interstitial content is voice broadcast, where the CPU mainly handles voice decoding and logic scheduling), then monitoring CPU utilization alone is sufficient. However, when the interstitial task involves the collaboration of multiple resources and the usage relationships are complex, a more comprehensive assessment is needed. For example, if the interstitial content is high-definition video or 3D animated advertisement, where the CPU is responsible for data parsing and task scheduling, and the GPU is responsible for image decoding and rendering, high utilization of either resource can lead to stuttering; therefore, joint monitoring of the CPU and GPU is required.

[0033] Specifically, system usage data can be directly obtained through system logs and the system control center. The current playback progress can also be directly read from the multimedia player. The historical insertion conflict rate is calculated by comparing the number of times of insertion stuttering, malfunctions, or other abnormal situations in the historical data within a historical time period with the total number of insertions.

[0034] Next, based on the insertion conflict assessment data, insertion feasibility calculation is performed to obtain an analytical value used to quantify the impact of insertion on system operation, denoted as the insertion conflict prediction coefficient. This value is then compared with the insertion conflict threshold used to determine whether an insertion conflict will occur. The insertion feasibility calculation involves normalizing each insertion conflict assessment data point (e.g., minimum-maximum normalization), multiplying each by an extracted preset influence factor, and then summing the results to obtain the corresponding insertion conflict prediction coefficient. Furthermore, these preset influence factors, used to reflect the degree of influence of each insertion conflict assessment data point on the insertion conflict prediction, are all set by preset professional staff, and the sum of all preset influence factors is 1. Similarly, the insertion conflict threshold is also data set by preset professional staff based on historical data and empirical rules, and is synchronously stored in a preset database for use.

[0035] It's important to note that before designing a machine learning-based in-vehicle information delivery decision-making system, technical professionals typically build a pre-set database to support various control strategies. This database integrates multiple key control parameters, including insertion context weights, preset influence factors, context relevance thresholds, delivery priority reduction magnitudes, priority adjustment ranges, and preset context weights. All parameters are pre-set by technical professionals based on the selected analysis method and on-site hardware configuration. This pre-set database forms the core data foundation for subsequent data uploading, storage optimization, and automated filtering and judgment processes.

[0036] If the prediction coefficient for delivery conflict is greater than the threshold for insertion conflict, it indicates that an insertion conflict has occurred. The deviation between the prediction coefficient for delivery conflict and the threshold for insertion conflict is mapped to obtain the insertion delay duration. After the insertion delay duration is reached, the insertion algorithm is activated to insert information. The insertion algorithm includes fade-in / fade-out algorithm and keyframe algorithm.

[0037] It should be added that the ratio of the difference between the prediction coefficient of the delivery conflict and the threshold for defining the insertion conflict to the threshold for defining the insertion conflict is recorded as the insertion conflict deviation. The insertion conflict deviation is input into the insertion conflict mapping table to obtain the corresponding insertion delay duration. This data table is used to fit the mapping relationship between the insertion conflict deviation and the insertion delay duration. Its construction method is as follows: in the initial data sequence constructed based on the linear regression algorithm, the insertion conflict deviation collected in the historical time period and the insertion delay duration set according to the empirical rules are used as training data, and the least squares method criterion and the statsmodels framework are used for training to finally obtain the trained insertion conflict mapping table.

[0038] If the prediction coefficient for delivery conflict is not greater than the threshold for insertion conflict, the insertion algorithm will be activated immediately for insertion.

[0039] Based on the above-mentioned conflict prediction and optimization, the conflict rate of insertion in vehicle information delivery decisions has been effectively reduced, such as... Figure 3 The figure shows a comparison diagram of the delivery conflict prediction optimization provided in the embodiment of this application. The figure shows the comparison of the insertion conflict rate before and after optimization over time. The horizontal axis is time (hours) and the vertical axis is the insertion conflict rate (%). The black curve represents the period before optimization and the red curve represents the period after optimization. The overall trend in the figure shows that the insertion conflict rate has a clear bimodal distribution throughout the day, appearing in the morning peak (approximately 7-9 am) and the evening peak (approximately 4-6 pm). This indicates that the insertion tasks are concentrated during commuting hours, resulting in a high system resource load. The overall conflict rate after optimization (red line) is lower than before optimization (black line), effectively reducing the probability of insertion conflicts. Especially during peak hours, the conflict rate drops by an average of about 3-5 percentage points. Moreover, the fluctuation range of the curve after optimization is significantly reduced, indicating that the system's insertion scheduling is more stable at different time periods, and the conflict situation is alleviated. In the early morning and late night periods (0-5 am and 21-24 pm), the conflict rate before and after optimization is close to 0%, indicating that the system has basically no conflicts when inserting under low load conditions. The optimization mainly plays a role during high load periods. This figure shows that by introducing a delivery conflict prediction and delay scheduling mechanism, the system can effectively allocate resources and reduce the conflict rate during high concurrency periods, achieving smooth and intelligent optimization of the insertion process, thereby significantly improving the stability of the vehicle information delivery decision system.

[0040] In this embodiment, by introducing a conflict prediction mechanism, intelligent control and system-level smooth scheduling of the vehicle-mounted information delivery decision-making system during the insertion process are achieved. Specifically, by acquiring system occupancy data and combining it with the current playback progress and historical insertion conflict rates, a more comprehensive assessment of the system's operating status is possible. This effectively avoids playback stuttering, task switching jitter, or task interruption problems caused by resource overload in traditional insertion methods, ensuring the stability of the information delivery process. Furthermore, by performing feasibility calculations on the insertion conflict assessment data and generating a conflict prediction coefficient, the system can quantify the impact of insertion operations on performance. The impact level is compared with the threshold for defining insertion conflicts, enabling dynamic judgment of insertion risk. Furthermore, when the analysis coefficient exceeds the threshold, the insertion delay duration is calculated through deviation mapping, giving the insertion execution adaptive buffering and flexible control capabilities. Insertion can be executed instantly when the system load is low, thus balancing system smoothness and real-time insertion. Moreover, the use of fade-in / fade-out and keyframe algorithms for insertion achieves smooth transitions between video and audio, improving visual and auditory continuity. Therefore, this solution is beneficial for improving the intelligent scheduling capability and information delivery stability of in-vehicle multimedia systems in complex operating environments.

[0041] It should be noted that the specific methods for AI-based context recognition and context-based delivery correlation assessment during the delivery conflict prediction process are as follows: Simultaneously, the AI ​​based on context recognition identifies the corresponding vehicle arrival context. If an insertion signal is issued based on context recognition AI, the correlation of the vehicle context insertion information is assessed, and the delivery priority of the insertion information is adjusted based on the corresponding assessment results. Delivery conflict prediction is then performed based on the adjusted delivery priority. Context recognition AI is used to determine the current external and operating environment of the vehicle, such as "location, weather, time, and driving status," in order to achieve contextualized delivery.

[0042] like Figure 4 The diagram shown is a flowchart of the vehicle context insertion information correlation evaluation provided in the embodiment of this application. The specific logic is: to obtain the context correlation index of the context recognition AI insertion information. The insertion context relevance index is weighted and fused with preset insertion context weights to obtain the corresponding insertion context relevance rating value. The insertion context relevance rating value is compared with the context relevance threshold: if the insertion context relevance rating value is higher than the context relevance threshold, the insertion information delivery priority is adjusted; otherwise, the delivery priority of the insertion information is not adjusted. Through the above process, it helps to improve the content relevance, information timeliness, and playback stability of the vehicle information delivery decision system.

[0043] The specific process for evaluating the relevance of vehicle-related contextualized broadcast information is as follows: The first step is to obtain the context relevance index of the AI-introduced information. The context relevance index includes semantic urgency and context relevance, and both of them have a value range of 0-1.

[0044] It's worth noting that several mature AI algorithms can directly calculate semantic urgency. For example, deep learning models (such as BERT, Bidirectional Encoder Representations from Transformers) can be used to identify urgent words (such as "alarm," "danger," "attention," and "urgent") in text. The model will output a standardized score (between 0 and 1), with higher values ​​indicating stronger semantic urgency. Contextual relevance can be obtained directly from existing systems or calculated by AI. For instance, the system can use an AI context matching model to fuse data such as GPS, weather, time, passenger flow, and vehicle sensor data to calculate the similarity between the inserted content and the current context. Common algorithms include: semantic similarity calculation based on vector embedding, machine learning models (such as random forests and XGBoost-eXtreme Gradient Boosting) to calculate matching scores based on multi-dimensional features, and deep learning context-awarerelevance models. Finally, a relevance score (also standardized to 0-1) is output, which is the contextual relevance.

[0045] The second step is to weight and fuse the insertion scenario relevance index with the preset insertion scenario weights to obtain the corresponding insertion scenario relevance rating. The insertion scenario relevance rating is used to reflect the relevance between the vehicle arrival scenario and the insertion information. The higher the relevance, the higher the priority of insertion delivery.

[0046] Specifically, the insertion context weights include semantic urgency weights and context relevance weights, which are preset data from professional staff. The sum of the two is 1. After multiplying each weight by semantic urgency and context relevance, the sum is obtained to obtain the corresponding insertion context relevance rating. In addition to the expert experience method mentioned above, the insertion context weights can also be set using statistical variance method and weight entropy method. The statistical variance method indicates that the weight is proportional to the feature variance, and the larger the variance, the higher the weight. The weight entropy method calculates the weight based on information entropy, and the smaller the entropy (the higher the degree of information dispersion), the greater the weight.

[0047] The third step is to compare the context relevance assessment value of the inserted message with the context relevance threshold used to define the relevance of the inserted message triggered by the current vehicle due to the context. The context relevance threshold is extracted from the preset database and is a preset value set by the preset staff.

[0048] Fourth, if the context relevance assessment value of the inserted message is higher than the context relevance threshold, the priority of the inserted message will be adjusted; otherwise, the priority of the inserted message will not be adjusted.

[0049] Specifically, the methods for adjusting the priority of ad insertion are as follows: The difference between the inserted context relevance rating and the context relevance threshold is called the context relevance difference value, which is the difference between the inserted context relevance rating and the context relevance threshold. The corresponding reduction in delivery priority is obtained by projecting the context relevance difference value.

[0050] It should be noted that the contextual relevance difference values ​​are input into the already trained contextual relevance mapping set, and the corresponding reduction in deployment priority is obtained by comparison. This mapping set is used to characterize the correlation between the contextual relevance difference values ​​and the reduction in deployment priority. Its construction process is as follows: the contextual relevance difference values ​​collected in the historical time period, and the reduction in deployment priority preset by professionals based on empirical rules, are input into the initial dataset constructed based on the logistic regression algorithm, and the least squares criterion is used in combination with the statsmodels framework to train the model, thereby obtaining the final contextual relevance mapping set.

[0051] The reduction in delivery priority is matched with the set priority adjustment range, that is, compared with the relationship of each range (belonging to or not belonging to the range), to obtain the corresponding reduction value of delivery priority.

[0052] The reduced placement priority value is added to the initial placement priority. This is done by adding the reduced placement priority value to the initial placement priority to obtain the adjusted placement priority. The second placement priority in the placement conflict prediction is then replaced with the adjusted placement priority for further analysis.

[0053] The priority adjustment intervals include, but are not limited to, priority level 1, priority level 2, and priority level 3 intervals, with the adjustment intensity increasing progressively at each level. The priority level 1 interval indicates that when the reduction in priority falls within this interval, the corresponding priority reduction value is one level, meaning the priority will be lowered by one level. The priority level 2 interval indicates that when the reduction in priority falls within this interval, the corresponding priority reduction value is two levels, meaning the priority will be lowered by two levels. The priority level 3 interval indicates that when the reduction in priority falls within this interval, the corresponding priority reduction value is three levels, meaning the priority will be lowered by three levels. Furthermore, the specific number of priority adjustment intervals is determined based on the specific information type of the in-vehicle multimedia system, and these intervals are generally preset by professional personnel based on historical data within a historical time period.

[0054] In this embodiment, by introducing an AI-based context recognition-based insertion context relevance analysis mechanism, intelligent priority control of in-vehicle multimedia information delivery is achieved. Specifically, by acquiring insertion context relevance indicators that include semantic urgency and contextual relevance, the system can comprehensively analyze the matching degree between the inserted content and the current vehicle operating context (such as location, weather, time, driving status, etc.) and the urgency of the information itself, thus making insertion decisions more in line with actual scenario needs. Secondly, by weighting and fusing each indicator with preset contextual weights, an insertion context relevance rating value is obtained, which not only achieves a comprehensive quantitative evaluation of multiple factors, but also makes the system more... The system can flexibly adjust weights according to operational strategies and achieve adaptive optimization under different application scenarios. Moreover, by comparing the evaluation value with the contextual relevance threshold, the system can automatically determine the necessity and priority of the inserted content. When the evaluation value is higher than the threshold, the insertion priority is automatically increased to ensure that urgent or highly relevant content is broadcast in a timely manner, while content below the threshold is executed as originally planned, thereby avoiding interference caused by frequent insertion of irrelevant information. Therefore, this solution realizes intelligent insertion decision based on AI contextual understanding, which effectively improves the content relevance, information timeliness, and playback stability of the in-vehicle information delivery decision system.

[0055] Example 2 To consider the delivery priority between inserted information and currently playing information, based on Example 1, delivery conflict prediction is performed, which also includes: Content-aware AI is used to acquire corresponding insertion information and identify the type of insertion information to obtain the corresponding delivery priority. The types of insertion information include, but are not limited to, emergency information, early warning information, commercial advertisements, and public service advertisements. Content-aware AI is used to identify the structure and scene changes of the currently playing content to determine the timing of insertion. The larger the value of the delivery priority, the lower the priority of the corresponding playback information. For example, the release priority of insertion information with a delivery priority of P0 is higher than that of playback information with a delivery priority of P1.

[0056] The current playback information is designated as the first priority, and the inserted information is designated as the second priority. If the first priority is greater than the second priority, a conflict prediction is performed on the inserted signal to determine the feasibility of insertion. If the first priority is not greater than the second priority, insertion is not allowed, the inserted information is cached in the current delivery task queue, and the insertion information is sorted by priority.

[0057] The specific details of prioritizing ad delivery are as follows: The priority of the inserted information is compared with the priority of each playback information in the delivery task queue, and the inserted information is inserted into the delivery task queue in order. If there is a case where the priority of the inserted information is the same as the priority of the playback information in the delivery task queue, the current inserted information is played first.

[0058] In this embodiment, by introducing a content-aware AI-based insertion priority determination mechanism, intelligent scheduling and orderly control of insertion tasks in the in-vehicle multimedia system are achieved. The system can automatically identify the type of inserted information (such as emergency broadcasts, advertisements, and prompts) and assign corresponding delivery priorities based on content characteristics, realizing intelligent and automated insertion decision-making. Furthermore, by comparing the priorities of the currently playing content and the inserted content, high-priority content can be inserted immediately, while low-priority content is cached and reordered, avoiding frequent interruptions to the playback process due to low-importance information, effectively ensuring the continuity of playback and the rational utilization of system resources. This solution makes information delivery more in line with the importance of the content and the needs of the audience, which is conducive to improving the information distribution efficiency of the in-vehicle multimedia system.

[0059] Example 3 To prevent interruptions in in-vehicle multimedia playback caused by the insertion delay, the insertion delay duration is determined based on Embodiment 1, and further includes: Read the remaining duration of the current playback information and compare it with the insertion delay duration. The comparison process is as follows: if the insertion delay duration is less than the remaining duration, the insertion will continue after the insertion delay duration; otherwise, the current insertion information will be cached in the delivery task queue and the insertion information will be played immediately after the current playback information finishes playing.

[0060] In this embodiment, an adaptive insertion scheduling mechanism based on a comparison between the remaining playback time and the insertion delay time is implemented to achieve the playback of in-vehicle multimedia information. Before inserting, the system detects the remaining playback time of the current information in real time and compares it with the insertion delay time to accurately determine the best trigger time for insertion. Secondly, when the insertion delay time is less than the remaining playback time, the system automatically executes the insertion after the delay, achieving timely delivery of the inserted content. When the insertion delay time is greater than the remaining playback time, the inserted content is automatically cached in the delivery task queue and switched immediately after the current content finishes playing, avoiding repeated interruptions or sudden changes in the screen. This mechanism takes into account both playback continuity and information timeliness, enabling the in-vehicle multimedia system to maintain a smooth and coordinated playback experience even in complex operating environments, effectively improving the system's intelligence level.

[0061] like Figure 5 The diagram shown is a flowchart illustrating the method of a machine learning-based vehicle information delivery decision system provided in this application, including: When the in-vehicle multimedia system is playing information according to the current delivery task queue, if an insertion signal is received, the system will predict the delivery conflict of the insertion signal to determine the feasibility of insertion, thereby ensuring the accuracy of the in-vehicle information delivery decision.

[0062] When predicting ad conflicts, AI is used to identify the context and assess the relevance of ad placements to determine whether to adjust the ad insertion data before ad conflict prediction.

[0063] After the interruption ends, playback resumes based on the task type of the current playback information. Task types include, but are not limited to, short interruptions, high-priority advertisements, emergency announcements, and multicast task overlays.

[0064] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)). Where there is no conflict, the solutions in the above embodiments can be combined.

[0065] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs) containing computer-usable program code. The form of a computer program product implemented on ROM, optical memory, etc.

[0066] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0067] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0069] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope and intent of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and variations.

Claims

1. A machine learning based in-vehicle information delivery decision system, characterized by, include: The content feature extraction module is used to identify the type of multimedia data through the first AI recognition model and generate corresponding delivery priority data. When the in-vehicle multimedia plays information in sequence, if an insertion signal is received, the module predicts the delivery conflict of the insertion signal to determine the feasibility of insertion. The context feature encoding module is used to predict delivery conflicts through the second AI decision model. When predicting delivery conflicts, the third AI context model is used to identify the current vehicle environment data. Based on AI, context identification and context delivery correlation assessment are performed to determine whether to adjust the insertion data before delivery conflict prediction. The strategy output module is used to restore the playback based on the task type of the current playback information after the interruption ends. The method of predicting delivery conflicts for inserted signals to determine the feasibility of insertion includes: Acquire insertion conflict assessment data, which includes system usage data, current playback content progress, and historical insertion conflict rate. The system usage data includes one or more of the following: CPU usage, GPU decoding load, network bandwidth usage, and memory usage. Based on the insertion conflict assessment data, insertion feasibility calculations are performed to obtain an analysis value used to quantify the impact of insertion on system operation. This analysis value is recorded as the insertion conflict prediction coefficient, and the insertion conflict prediction coefficient is compared with the insertion conflict definition threshold used to determine whether an insertion conflict will occur. Among them, the insertion feasibility calculation means that after normalizing the data of each insertion conflict assessment, it is multiplied by the extracted preset impact factor, and then added to obtain the corresponding insertion conflict prediction coefficient. If the delivery conflict prediction coefficient is greater than the insertion conflict threshold, it indicates that an insertion conflict has occurred. The deviation between the delivery conflict prediction coefficient and the insertion conflict threshold is mapped to obtain the insertion delay duration. After the insertion delay duration is reached, the insertion algorithm is activated to insert the insertion information. The insertion algorithm includes a fade-in / fade-out algorithm and a keyframe algorithm. If the prediction coefficient of the delivery conflict is not greater than the threshold for the insertion conflict, the insertion algorithm will be activated immediately for insertion. The ratio of the difference between the prediction coefficient of the delivery conflict and the threshold for defining the insertion conflict to the threshold for defining the insertion conflict is recorded as the insertion conflict deviation. The insertion conflict deviation is input into the insertion conflict mapping table to obtain the corresponding insertion delay duration. This insertion conflict mapping table is used to fit the mapping relationship between the insertion conflict deviation and the insertion delay duration. Its construction method is as follows: In the initial data sequence constructed based on the linear regression algorithm, the insertion conflict deviation collected in the historical time period and the insertion delay duration set according to the empirical rules are used as training data, and the least squares method criterion and the statsmodels framework are used for training to finally obtain the trained insertion conflict mapping table. The process of obtaining the insertion delay duration further includes: The remaining duration of the current playback information is read and compared with the insertion delay duration. The comparison process is as follows: if the insertion delay duration of the insertion information is less than the remaining duration, the insertion information is continued to be inserted after the insertion delay duration; otherwise, the corresponding insertion information is cached in the delivery task queue and the insertion information is played immediately after the current playback information finishes playing.

2. The in-vehicle information delivery decision system based on machine learning as described in claim 1, characterized in that: The process of predicting delivery conflicts also includes, prior to: Based on content-aware AI, the insertion information corresponding to the insertion signal is obtained, and the type of the insertion information is identified to obtain the corresponding delivery priority. The larger the value of the delivery priority, the lower the priority of the corresponding insertion information. The current playback information is designated as the first priority, and the inserted information is designated as the second priority. If the first delivery priority is greater than the second delivery priority, then delivery conflict prediction is performed on the insertion signal to determine the feasibility of insertion; If the first delivery priority is not greater than the second delivery priority, then no insertion is allowed. The insertion information is cached in the current delivery task queue and the insertion information is sorted by delivery priority.

3. The in-vehicle information delivery decision system based on machine learning as described in claim 2, characterized in that: The process of prioritizing delivery includes: The priority of the inserted information is compared with the priority of each playback information in the delivery task queue. Based on the comparison result, the inserted information is inserted into the delivery task queue in order. If there is a case where the priority of the inserted information is the same as the priority of the playback information in the delivery task queue, the current inserted information is played first.

4. The in-vehicle information delivery decision system based on machine learning as described in claim 1, characterized in that: The specific methods for AI-based context recognition and context-based adjudication relevance assessment are as follows: During the process of predicting delivery conflicts, the corresponding vehicle arrival context is simultaneously identified based on context recognition AI. If the AI-based context recognition system releases the insertion signal, then the correlation between the insertion information and the vehicle context is assessed. Based on the corresponding evaluation results, the priority of the inserted information is adjusted, and the conflict prediction is carried out according to the adjusted priority.

5. The in-vehicle information delivery decision system based on machine learning as described in claim 4, characterized in that: The assessment of the relevance of the vehicle contextualized information includes: Obtain the context relevance index of AI-powered insertion information for context recognition, wherein the context relevance index includes semantic urgency and context relevance; The insertion context relevance index is weighted and fused with the preset insertion context weights to obtain the corresponding insertion context relevance rating value. The higher the insertion context relevance rating value, the higher the insertion delivery priority. The context relevance rating of the inserted message is compared with the context relevance threshold used to define the relevance of the inserted message triggered by the current vehicle due to a specific context: If the context relevance rating of the inserted message is higher than the context relevance threshold, the priority of the inserted message will be adjusted; otherwise, the priority of the inserted message will not be adjusted.

6. The in-vehicle information delivery decision system based on machine learning as described in claim 5, characterized in that: The adjustment of the priority of the inserted broadcast includes: The difference between the context relevance assessment value and the context relevance threshold is recorded as the context relevance difference value. The corresponding reduction in delivery priority is obtained by projecting the context relevance difference value. The reduction in delivery priority is matched with the set priority adjustment range to obtain the corresponding reduction value in delivery priority; The reduced delivery priority value is superimposed with the initial delivery priority to obtain the adjusted delivery priority. The second delivery priority in the delivery conflict prediction is replaced with the adjusted delivery priority, and the inserted information is analyzed again.

7. The in-vehicle information delivery decision system based on machine learning as described in claim 6, characterized in that: The priority adjustment range includes a first-level priority adjustment range, a second-level priority adjustment range, and a third-level priority adjustment range, which represent a progressively increasing adjustment intensity of the deployment priority. The first-level priority adjustment range indicates that when the decrease in the priority of the delivery falls within this range, the corresponding delivery priority is reduced by one level. The priority level two adjustment range means that when the decrease in the priority of the delivery falls within this range, the corresponding delivery priority is reduced by two levels. The three-level priority adjustment range means that when the reduction in the priority of the delivery falls within this range, the corresponding priority of the delivery will be reduced by three levels.

8. A machine learning-based decision-making method for vehicle information delivery, characterized in that, The method employs the machine learning-based vehicle information delivery decision system as described in any one of claims 1-7, and the method includes: When the in-vehicle multimedia system is playing information according to the current playback queue order, if an insertion signal is received, the insertion signal is subjected to a delivery conflict prediction to determine the feasibility of insertion. When predicting ad conflicts, AI is used to identify the context and assess the relevance of the context to determine whether to adjust the ad insertion data before ad conflict prediction. After the interruption ends, resume playback based on the task type of the current playback information.

Images