Streaming media playing-based expansible content interactive display method and system

Abstract

The invention discloses an extensible content interactive display method and system based on streaming media playing, the extensible content interactive display method based on streaming media playing belongs to the field of streaming media, and the method comprises the following steps: S1, dynamically adjusting the resolution and code rate of content under different network conditions; dynamic self-adaptive stream transmission is realized; s2, directly acquiring data through the nearest edge node when the content is requested, and caching high-frequency interaction data at the edge node; s3, the user request is routed to the optimal node; s4, adopting a predictive buffering and preloading mechanism; s5, predicting user interaction behaviors by using a deep reinforcement learning algorithm; and S6, adopting H.265, and combining the characteristics of low delay and high reliability of the QUIC protocol to carry out transmission. The extensible content interactive display system based on streaming media playing comprises a user behavior analysis module, a content management and distribution module, an interactive prediction and recommendation module, a preloading and cache management module, a network transmission and optimization module and the like.

Description

Technical Field

[0001] The present invention relates to the field of streaming media, and more specifically, to an extensible content interactive display method and system based on streaming media playback. Background Art

[0002] The scalable content interactive display method and system for streaming media playback is an interactive experience method based on real-time data transmission and dynamic content generation. The benefits of this method are mainly reflected in several aspects. First, it enhances the sense of participation in the user experience. By allowing users to interact during the playback process, users can directly affect the presentation and progress of the content. For example, viewers can choose different plot branches to obtain a customized plot experience, which greatly enhances the personalization and interactivity of the content. Secondly, the real-time and flexibility of streaming media enables the interactive display content to be continuously updated, and creators can adjust the content in real time based on audience feedback. Furthermore, the interactive display method based on streaming media can integrate different multimedia elements, such as audio, video, text, animation, etc., to make the content expression more rich and diverse. In addition, this method usually adopts cloud computing and distributed architecture, which can be effectively expanded to meet the synchronous interaction needs of large-scale users and enhance the dissemination and influence of content.

[0003] However, despite its many advantages, the biggest drawback of the streaming interactive display method is its high dependence on network bandwidth. To ensure a real-time interactive experience, streaming content needs to continuously transmit data, and when the number of users increases, the network load will increase, increasing bandwidth pressure, especially when displaying high-definition or ultra-high-definition content. If the network condition is poor or the bandwidth is insufficient, the user experience may be greatly reduced, and problems such as screen freezes and delays may occur, thereby weakening the interactive experience. Summary of the invention

[0004] The purpose of the present invention is to provide an extensible content interactive display method and system based on streaming media playback to solve the problems raised in the above background technology: to ensure a real-time interactive experience, streaming media content needs to continuously transmit data, and when the number of users increases, the network load will increase, and the bandwidth pressure will increase, especially in the case of high-definition or ultra-high-definition content display. If the network condition is poor or the bandwidth is insufficient, the user experience may be greatly reduced, and problems such as screen freeze and delay may occur, thereby weakening the interactive experience effect.

[0005] Technical solution: The method for interactive display of extensible content based on streaming media playback includes the following steps:

[0006] S1. Based on multi-bitrate encoding and dynamic adaptive streaming protocol, dynamically adjust the resolution and bit rate of the content under different network conditions; use ABR technology to detect network bandwidth, combine the dynamic adaptive streaming protocol to automatically switch the bit rate on the client side, and realize dynamic adaptive streaming transmission;

[0007] S2. Use edge computing architectures such as Amazon CloudFront and Akamai Edge to cache and distribute data at CDN edge nodes. When requesting content, directly obtain data through the nearest edge node and cache high-frequency interactive data at the edge node.

[0008] S3. Based on the Amazon CloudFront and Akamai Edge, use multiple nodes around the world to distribute content; use DNS load balancing and intelligent routing strategies to route user requests to the optimal node, and use CDN cache to accelerate transmission;

[0009] S4. Use predictive buffering and preloading mechanisms, combined with AI sequence models to predict interactive content, use sequence models to cache predicted content in advance, and use HTTP / 2 Server Push to preload user interactive content;

[0010] S5. Apply deep reinforcement learning algorithm to predict user interaction behavior and pre-fetch in the background; pre-load predicted content in the background through the HTTP / 2ServerPus subscription protocol;

[0011] S6. Utilize the high-efficiency video coding format H.265 and the QUIC transmission protocol to optimize streaming media transmission efficiency, and use the H.265 in combination with the low latency and high reliability characteristics of the QUIC protocol for transmission.

[0012] Preferably, the S2 introduces a content-aware dynamic caching strategy, which analyzes user viewing preferences and content popularity, and preferentially caches high-frequency data and key interactive content that users interact with most frequently to edge nodes.

[0013] Preferably, the specific implementation steps are as follows:

[0014] S2-1. Use machine learning algorithms to analyze user viewing behavior, identify content with high interaction frequency and popularity, and give priority to it; combine user historical behavior and overall popularity trends to prioritize video clips;

[0015] S2-2. Based on the priority, the cache allocation strategy is dynamically adjusted through the adaptive allocation algorithm to cache high-priority content, including key interactive nodes and high-frequency fragments, to edge nodes;

[0016] S2-3. Use LRU and LFU hybrid cache replacement strategy; cache replacement for low priority content;

[0017] S2-4. Real-time monitoring of edge node bandwidth usage, dynamic adjustment of cache content resolution and transmission bit rate, and encoding of low-priority content with higher compression rates;

[0018] S2-5. Optimize back-to-source data transmission by regularly synchronizing user data and interaction records, and quickly obtain content from the back-to-source server when the content exceeds the capacity of the edge node and the priority changes.

[0019] Preferably, S2-1 uses the clustering algorithm K-Means to classify the interactive behavior data and identify high-frequency interactive scenes; quantify the number of clicks and interactive time of each interactive node, and label the content with high or low priority according to the frequency, and identify "high-frequency interactive" content by analyzing the interactive frequency clustering results, including important plot nodes and interactive selection points where users are concentrated;

[0020] The time series model RIMA is used to predict changes in content popularity; the future popularity of the content is predicted by learning the interaction data trends over the past period of time; the recent interaction data is input into the time series model in real time to dynamically update the popularity prediction results so that the content priority reflects the current user interests.

[0021] Preferably, the calculation formula for the clustering algorithm K-Means to classify the interactive behavior data and identify high-frequency interactive scenes is as follows:

[0022] Initialize cluster centers Randomly select K initial cluster centers, calculate distances and assign data points to the nearest cluster center. For each interactive behavior data point x i , calculate its difference with all cluster centers μ j The Euclidean distance of:

[0023]

[0024] Where n is the number of feature dimensions, x ik is the data point x i The value of the K-th feature, μ jk is the cluster center μ j The value of the K-th feature;

[0025] Assign all data points to the cluster with the closest distance to obtain K clusters; then, update the μ of each cluster center j , calculate the mean of all data points belonging to this cluster:

[0026]

[0027] Among them, the C j represents the set of data points in cluster j, the |C j | is the number of data points in cluster j;

[0028] Repeat steps 2 and 3 until the position of the cluster center no longer changes significantly and the preset upper limit of the number of iterations is reached;

[0029] After clustering, analyze the feature center of each cluster; cluster centers with higher interaction frequency and dwell time can be identified as "high-frequency interaction scenes", and the central feature values ​​of these clusters represent the user's high interaction intention; define a threshold T to determine whether a cluster belongs to a high-frequency interaction scene; for a cluster j, if the number of clicks and dwell time of its feature center are both higher than the threshold T, then the cluster is defined as a high-frequency interaction scene cluster.

[0030] Preferably, the S2-1 uses a dynamic partitioning algorithm to adjust the cache capacity based on priority scoring, places high-priority content in the CDN, combines the Least Recently Used and Least Frequently Used mixed replacement strategies, and uses the user clustering algorithm K-Means to group users, identify groups with different viewing behaviors, and distinguish users who prefer plot interaction from users who like scene exploration; according to the preferences of user groups, differentiated content priorities are assigned to different groups; the plot interaction preference group will give priority to caching core interactive content, while the exploration group will give priority to caching scene content.

[0031] Preferably, the comprehensive priority score of the content is calculated by a weighted formula, and the calculation formula is as follows:

[0032] Priority score = w1 × interaction frequency + w2 × content popularity

[0033] Among them, w1 and w2 represent the weights of interaction frequency and content popularity respectively, and are dynamically adjusted according to business needs; the content is sorted according to the priority score; high-scoring content is marked as priority cache content, and low-scoring content is downgraded to non-priority cache content.

[0034] Preferably, the S5 deep reinforcement learning algorithm predicts user interaction behavior including the following steps:

[0035] Define the user's current viewing status, including current content type, selected interaction path, viewing time, and interaction frequency characteristics;

[0036] Action (A): The interactive operation that the user takes in the next step, such as clicking on specific content or selecting an interactive option;

[0037] Reward (R): Design a reward mechanism for the accuracy of user behavior prediction; if the interaction content predicted by the system is consistent with the user's actual choice, a positive reward is given; otherwise, a negative reward is given; this mechanism guides the model to continuously optimize the prediction effect;

[0038] Use a deep neural network to build a Q network model, input the user's current state characteristics, and output the Q value of each interactive operation, that is, the expected reward value of each action. The Q network formula is as follows:

[0039]

[0040] Wherein, s represents the current state, a represents the current action, r represents the reward value, s′ and a′ represent the next state and the optimal action respectively, and γ is the discount factor;

[0041] The deep neural network is used to build a Q network model, the user's current state features are input, and the Q value of each interactive operation, that is, the expected reward value of each action, is output; the collected user historical behavior data is used to build a training set, the playback mechanism is used to improve the model stability, and the target network parameters are updated regularly; in an actual environment, the user's newly generated behavior data is collected in real time, and the collected behavior data is input into the model for fine-tuning.

[0042] Preferably, the predicted content is preloaded using the HTTP / 2 ServerPush protocol, comprising the following steps:

[0043] S5-1. Set a push list and use the high-probability content predicted by the Q network model as a push resource; once the server detects that the user starts loading a video clip, it immediately pushes the predicted content to the client according to the push list;

[0044] S5-2. According to the predicted probability of the Q network model, different priorities are set for the pushed content; high-probability content is set to high priority and preloaded first; combined with the client cache strategy, ensure that the predicted content pushed is cached in the user device, and low-priority content is dynamically loaded after the user selects it;

[0045] S5-3. Configure a loading control mechanism on the client to allow users to directly load the next predicted interactive content without delay when completing the current interaction; when the user's actual selection does not match the pushed content, the system will first extract the content from the cache; if the content does not exist in the cache, it will be loaded from the server immediately;

[0046] S5-4. Update the reward mechanism of the reinforcement learning model by comparing the user's actual choice with the predicted result; regularly perform model training iterations and input the latest user behavior data into the Q network model.

[0047] Preferably, the extensible content interactive display system based on streaming media playback includes rights 1-9 of the extensible content interactive display method based on streaming media playback, including a user behavior analysis module, a content management and distribution module, an interactive prediction and recommendation module, a preloading and cache management module, a network transmission and optimization module, a content display and rendering module, a data collection and feedback module, and a security and authority management module.

[0048] Compared with the prior art, the advantages of the present invention are:

[0049] (1) Use deep reinforcement learning algorithms to predict user interaction behaviors and analyze user behavior data in real time. This enables the system to identify user preferences in advance and predict future interaction choices more accurately, rather than relying solely on simple behavioral statistics or fixed logic.

[0050] (2) Based on user behavior prediction, the system preloads high-probability interactive content through the HTTP / 2 Server Push protocol and actively pushes the predicted content to the client, thereby reducing loading delays during user interaction. This preloading strategy not only improves the user experience, but also reduces the load of requesting data after the user makes a selection.

[0051] (3) The system calculates the interaction frequency and popularity of the content and assigns dynamic priority to different content, ensuring that content with high interaction frequency and high popularity is cached first. This ensures fast access to high-demand content when resources are limited, thereby improving the efficiency of content distribution and the smoothness of user experience.

[0052] (4) Through intelligent preloading protocol and dynamic buffer window technology, the system can maintain a smooth interactive experience even when the network bandwidth is unstable. This design is more adaptable and helps improve user satisfaction under different network conditions.

[0053] (5) The system uses CDN edge nodes to distribute high-priority content and reduce the transmission delay of remote servers. In addition, user behavior clustering analysis and personalized caching strategies enable the system to optimize content push according to the characteristics of user groups, achieving more accurate and real-time interactive services.

[0054] (6) The system includes a data feedback and model iteration module that can automatically update the behavior prediction model, continuously improve prediction accuracy, and optimize cache management; this closed-loop mechanism ensures that the system can dynamically adapt to changes in user behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the overall system of the extensible content interactive display method based on streaming media playback of the present invention;

[0056] Figure 2 It is an overall system diagram of the extensible content interactive display method based on streaming media playback of the present invention. DETAILED DESCRIPTION

[0057] For examples, see Figure 1 , an extensible content interactive display method based on streaming media playback, the extensible content interactive display method based on streaming media playback comprises the following steps:

[0058] S1. Based on multi-bitrate encoding and dynamic adaptive streaming protocol, the resolution and bit rate of the content are dynamically adjusted under different network conditions; ABR technology is used to detect network bandwidth, and the bit rate is automatically switched on the client side in combination with the dynamic adaptive streaming protocol to achieve dynamic adaptive streaming transmission;

[0059] Specifically, the content classification and compression optimization operation steps are as follows;

[0060] a. Adaptive Bitrate Coding (ABR) Configuration

[0061] Use Adaptive Bitrate Streaming (ABR) technology to configure different resolutions and bitrate versions for video streams, such as 1080p, 720p, 480p, etc.

[0062] Each version is compressed at a different ratio to ensure clear, continuous images even in low bandwidth situations.

[0063] B. Real-time bandwidth monitoring

[0064] During video playback, the network bandwidth and latency are detected in real time through the client device.

[0065] Based on the current bandwidth status, the system selects the most suitable streaming resolution and bitrate version to ensure that the video can be played continuously without interruption or buffering.

[0066] This process can be optimized through bandwidth estimation models (such as buffer mode in the ABR algorithm, network condition model, etc.).

[0067] c. Content classification and pre-setting strategy

[0068] Video content is classified into different quality levels based on content importance, interaction needs, and user preferences.

[0069] Core interactive points or important plots use higher resolutions, while other non-core parts use lower resolutions to achieve a balance between bandwidth saving and content display quality.

[0070] E. Compression technology optimization

[0071] Use high-efficiency video coding formats such as H.265 (HEVC) or AV1, which can reduce the amount of data by about 30-50% compared to traditional H.264 at the same image quality.

[0072] Using video content-aware encoding (CAE) technology, low-detail areas can be compressed more strongly without affecting the viewing experience, while maintaining high encoding quality for high-detail areas.

[0073] F. Dynamic decoding of edge devices

[0074] Supports hardware-accelerated decoding on edge devices, such as using GPU-accelerated hardware decoding on mobile devices and smart TVs, ensuring smooth playback of highly compressed videos even under low bandwidth.

[0075] Use video decoding algorithms to make real-time frame rate adjustments on the device side based on network fluctuations to alleviate network load.

[0076] G. Intelligent caching and preloading

[0077] Cache high-quality video clips in advance when users are about to enter interactive points or areas with high bandwidth demands.

[0078] Use server-side prediction and client-side preloading to reduce waiting time when interactive points are loaded.

[0079] H. Adaptive Feedback and Model Optimization

[0080] Continuously monitor user experience and bandwidth consumption, and use the feedback mechanism of the ABR algorithm to adjust the bitrate selection.

[0081] Use machine learning algorithms to analyze users’ viewing habits and network conditions, optimize the adaptive streaming model, and further reduce bandwidth pressure.

[0082] S2. Use edge computing architectures such as Amazon CloudFront and Akamai Edge to cache and distribute data at CDN edge nodes. When requesting content, directly obtain data through the nearest edge node and cache high-frequency interactive data at the edge node.

[0083] S3. Based on Amazon CloudFront and Akamai Edge, it uses multiple nodes around the world to distribute content. Through DNS load balancing and intelligent routing strategies, it routes user requests to the optimal node and uses CDN cache to accelerate transmission.

[0084] S4. Use predictive buffering and preloading mechanisms, combined with AI sequence models to predict interactive content, use sequence models to cache predicted content in advance, and use HTTP / 2 Server Push to preload user interactive content;

[0085] Specifically, the specific operation includes the following steps:

[0086] 1. Predictive caching and preloading mechanisms

[0087] The core of predictive buffering and preloading is to load the content that users may need into the client's cache in advance, ensuring that the system has prepared content resources when users interact. This approach can avoid delays caused by remote loading of content. To achieve this, the system uses an intelligent content prediction strategy that uses AI sequence models to analyze and judge users' viewing behavior and interaction trends, thereby determining which content needs to be preloaded.

[0088] 2. Interactive content prediction combined with AI sequence model

[0089] The system uses AI sequence models (such as recurrent neural network RNN, long short-term memory network LSTM or variant model Transformer) to learn and predict user interaction sequences. The specific implementation process is as follows:

[0090] User behavior modeling: The model uses data such as each step of the user's operation, viewing time, interaction frequency, etc. as input features to form a time series that records the user's viewing and interaction habits.

[0091] Predict the next action: Through learning the sequence model, the system can generate the probability distribution of the user's next interactive content, thereby determining the next piece of content that the user is most likely to choose. The sequence model continuously learns the user's behavior patterns to improve the accuracy of the prediction.

[0092] Dynamically update the model: The model is constantly updated in real time to adapt to the user's personalized needs and changing trends. This allows the system to quickly adjust the prediction content when user behavior changes.

[0093] 3. Cache predicted content in advance

[0094] Using the content predicted by the AI ​​sequence model, the system caches the high-probability content as "predicted content" in advance to the user's device. The specific steps are as follows:

[0095] Determine content priority: Based on the predicted probability of the AI ​​model, prioritize the content that may be selected by the user next, and assign high priority to the content that is most likely to be accessed.

[0096] Cache strategy implementation: The system caches content resources in different levels according to content priority. High-priority content will be cached immediately and ready to be loaded to users at any time, while low-priority content can be dynamically loaded when needed.

[0097] Cache monitoring and replacement: When cache space is limited, the system will clear low-priority content based on priority to ensure that the content in the cache is always the content that the user is most likely to need.

[0098] 4. Use HTTP / 2 Server Push to load interactive content in advance

[0099] Once the predicted content is determined, the system pushes the content to the client through the HTTP / 2 ServerPush protocol so that it can be loaded before the user makes an actual request. The implementation steps include:

[0100] Push list configuration: The system generates a push list that packages the most likely interactive options in the predicted content. When the user loads the current interactive content, the system pushes the content that may be accessed next to the client through ServerPush based on the list.

[0101] Client caching and preloading: Content pushed to the client is automatically loaded into the browser cache. This way, when the user makes the next choice, the content can be loaded directly from the local computer, avoiding the waiting time caused by network delays.

[0102] Dynamically adjust push content: Based on real-time changes in user behavior, the system will update the push list to ensure that the push content always matches the user's interests and habits. The HTTP / 2 Server Push mechanism allows the system to flexibly adjust push content, further improving preloading efficiency.

[0103] 5. Optimize network resource utilization and improve user experience

[0104] Through predictive caching and preloading mechanisms, the system can optimize the use of network resources while ensuring user experience:

[0105] Bandwidth saving: The preloading strategy avoids users from repeatedly loading content and reduces ineffective bandwidth consumption.

[0106] Improved user experience: Every interaction of the user can be responded to instantly, because the content has been loaded to the client in advance, there is no need to wait for loading, which significantly improves the smoothness of streaming media interaction.

[0107] S5. Apply deep reinforcement learning algorithms to predict user interaction behaviors and pre-fetch in the background; pre-load predicted content in the background through HTTP / 2ServerPus subscription protocol;

[0108] S6. Use the high-efficiency video coding format H.265 and the QUIC transmission protocol to optimize streaming media transmission efficiency, and use H.265 combined with the low latency and high reliability characteristics of the QUIC protocol for transmission.

[0109] S2 introduces a content-aware dynamic caching strategy that analyzes user viewing preferences and content popularity, and prioritizes caching high-frequency data and key interactive content that users interact with most frequently to edge nodes.

[0110] The specific implementation steps are as follows:

[0111] S2-1. Use machine learning algorithms to analyze user viewing behavior, identify content with high interaction frequency and popularity, and give priority to it; combine user historical behavior and overall popularity trends to prioritize video clips;

[0112] S2-2. Based on the priority, the cache allocation strategy is dynamically adjusted through the adaptive allocation algorithm to cache high-priority content, including key interactive nodes and high-frequency fragments, to edge nodes;

[0113] S2-3. Use LRU and LFU hybrid cache replacement strategy; cache replacement for low priority content;

[0114] S2-4. Real-time monitoring of edge node bandwidth usage, dynamic adjustment of cache content resolution and transmission bit rate, and encoding of low-priority content with higher compression rates;

[0115] S2-5. Optimize back-to-source data transmission by regularly synchronizing user data and interaction records, and quickly obtain content from the back-to-source server when the content exceeds the capacity of the edge node and the priority changes.

[0116] S2-1 uses the clustering algorithm K-Means to classify the interactive behavior data and identify high-frequency interactive scenarios; quantifies the number of clicks and interaction time of each interactive node, and labels the content with high or low priority based on the frequency. By analyzing the clustering results of the interaction frequency, "high-frequency interactive" content is identified, including important plot nodes and interactive selection points where users are concentrated;

[0117] The time series model RIMA is used to predict changes in content popularity. The future popularity of content is predicted by learning the interaction data trends over the past period of time. The recent interaction data is input into the time series model in real time to dynamically update the popularity prediction results so that the content priority reflects the current user interests.

[0118] The calculation formula of the clustering algorithm K-Means to classify the interactive behavior data and identify high-frequency interactive scenes is as follows:

[0119] Initialize cluster centers Randomly select K initial cluster centers, calculate distances and assign data points to the nearest cluster center. For each interactive behavior data point x i , calculate its difference with all cluster centers μ j The Euclidean distance of:

[0120]

[0121] Where n is the number of feature dimensions, x ik is the data point x i The value of the Kth feature, μ jk is the cluster center μ j The value of the Kth feature;

[0122] Assign all data points to the cluster with the closest distance to obtain K clusters; then, update the center μ of each cluster j , calculate the mean of all data points belonging to this cluster:

[0123]

[0124] Among them, C j represents the set of data points in cluster j, |C j | is the number of data points in cluster j;

[0125] Repeat steps 2 and 3 until the position of the cluster center no longer changes significantly and the preset upper limit of the number of iterations is reached;

[0126] After clustering, analyze the feature center of each cluster; cluster centers with higher interaction frequency and dwell time can be identified as "high-frequency interaction scenarios", and the central feature values ​​of these clusters represent the user's high interaction intention; define a threshold T to determine whether a cluster belongs to a high-frequency interaction scenario; for a cluster j, if the number of clicks and dwell time of its feature center are both higher than the threshold T, then the cluster is defined as a high-frequency interaction scenario cluster.

[0127] S2-1 uses a dynamic partitioning algorithm to adjust cache capacity based on priority scoring, places high-priority content in the CDN, combines the Least Recently Used and Least Frequently Used hybrid replacement strategies, and uses the user clustering algorithm K-Means to group users, identify groups with different viewing behaviors, and distinguish users who prefer plot interaction from users who prefer scene exploration; according to the preferences of user groups, differentiated content priorities are assigned to different groups; the plot interaction preference group will prioritize caching core interactive content, while the exploration group will prioritize caching scene content.

[0128] The comprehensive priority score of the content is calculated using a weighted formula, which is as follows:

[0129] Priority score = w1 × interaction frequency + w2 × content popularity

[0130] Among them, w1 and w2 represent the weights of interaction frequency and content popularity respectively, which are dynamically adjusted according to business needs; the content is sorted according to the priority score; high-scoring content is marked as priority cache content, and low-scoring content is downgraded to non-priority cache content.

[0131] The S5 deep reinforcement learning algorithm predicts user interaction behavior by following these steps:

[0132] Define the user's current viewing status, including current content type, selected interaction path, viewing time, and interaction frequency characteristics;

[0133] Use a deep neural network to build a Q network model, input the user's current state characteristics, and output the Q value of each interactive operation, that is, the expected reward value of each action. The Q network formula is as follows:

[0134]

[0135] Where s represents the current state, a represents the current action, r represents the reward value, s′ and a′ represent the next state and the optimal action respectively, and γ is the discount factor;

[0136] Use a deep neural network to build a Q network model, input the user's current state features, and output the Q value of each interactive operation, that is, the expected reward value of each action; use the collected user historical behavior data to build a training set, use the playback mechanism to improve the model stability, and regularly update the target network parameters; in the actual environment, collect new user behavior data in real time, and input the collected behavior data into the model for fine-tuning.

[0137] Preloading predicted content using the HTTP / 2 ServerPush protocol includes the following steps:

[0138] S5-1. Set a push list and use the high-probability content predicted by the Q network model as a push resource; once the server detects that the user starts loading a video clip, it immediately pushes the predicted content to the client according to the push list;

[0139] S5-2. According to the prediction probability of the Q network model, different priorities are set for the pushed content; high-probability content is set as high priority and preloaded first; combined with the client cache strategy, ensure that the pushed predicted content is cached in the user's device, and low-priority content is dynamically loaded after the user selects it;

[0140] S5-3. Configure a loading control mechanism on the client to allow users to directly load the next predicted interactive content without delay when completing the current interaction; when the user's actual selection does not match the pushed content, the system will first extract the content from the cache; if the content does not exist in the cache, it will be loaded from the server immediately;

[0141] S5-4. Update the reward mechanism of the reinforcement learning model by comparing the user's actual choices with the predicted results; regularly perform model training iterations and input the latest user behavior data into the Q network model.

[0142] The extensible content interactive display system based on streaming media playback includes rights 1-9 of the extensible content interactive display method based on streaming media playback, including a user behavior analysis module, a content management and distribution module, an interactive prediction and recommendation module, a preloading and cache management module, a network transmission and optimization module, a content display and rendering module, a data collection and feedback module and a security and permission management module.

[0143] The above shows and describes the basic principles, main features and advantages of the present invention; those skilled in the art should understand that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected; the scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. An extensible content interactive display method based on streaming media playback, characterized in that: The method for interactively displaying extensible content based on streaming media playback comprises the following steps: S1. Based on multi-bitrate encoding and dynamic adaptive streaming protocol, dynamically adjust the resolution and bit rate of the content under different network conditions; use ABR technology to detect network bandwidth, combine the dynamic adaptive streaming protocol to automatically switch the bit rate on the client side, and realize dynamic adaptive streaming transmission; S2. Use edge computing architectures such as Amazon CloudFront and Akamai Edge to cache and distribute data at CDN edge nodes; directly obtain data through the nearest edge node when requesting content, and cache high-frequency interactive data at the edge node; S3. Based on the Amazon CloudFront and Akamai Edge, use multiple nodes around the world to distribute content; use DNS load balancing and intelligent routing strategies to route user requests to the optimal node, and use CDN cache to accelerate transmission; S4. Use predictive buffering and preloading mechanisms, combined with AI sequence models to predict interactive content, use the AI ​​sequence model to cache predicted content in advance, and use HTTP / 2 Server Push to load user interactive content in advance; S5. Apply deep reinforcement learning algorithm to predict user interaction behavior and pre-fetch in the background; pre-load predicted content in the background through the HTTP / 2ServerPus subscription protocol; S6. Utilize the high-efficiency video coding format H.265 and the QUIC transmission protocol to optimize streaming media transmission efficiency, and use the H.265 in combination with the low latency and high reliability characteristics of the QUIC protocol for transmission.

2. The method for interactively displaying extensible content based on streaming media playback according to claim 1, characterized in that: The S2 introduces a content-aware dynamic caching strategy, which analyzes user viewing preferences and content popularity, and prioritizes caching high-frequency data and key interactive content that users interact with most frequently to edge nodes.

3. The method for interactively displaying extensible content based on streaming media playback according to claim 2, characterized in that: The specific implementation steps are as follows: S2-1. Use machine learning algorithms to analyze user viewing behavior, identify content with high interaction frequency and popularity, and give priority to it; combine user historical behavior and overall popularity trends to prioritize video clips; S2-2. Based on the priority, the cache allocation strategy is dynamically adjusted through the adaptive allocation algorithm to cache high-priority content, including key interactive nodes and high-frequency fragments, to edge nodes; S2-3. Use LRU and LFU hybrid cache replacement strategy; cache replacement for low priority content; S2-4. Real-time monitoring of edge node bandwidth usage, dynamic adjustment of cache content resolution and transmission bit rate, and encoding of low-priority content with higher compression rates; S2-5. Optimize back-to-source data transmission by regularly synchronizing user data and interaction records, and quickly obtain content from the back-to-source server when the content exceeds the capacity of the edge node and the priority changes.

4. The method for interactively displaying extensible content based on streaming media playback according to claim 3 is characterized in that: S2-1 uses the clustering algorithm K-Means to classify the interactive behavior data and identify high-frequency interactive scenes; quantifies the number of clicks and interaction time of each interactive node, and labels the content with high or low priority according to the frequency. By analyzing the interactive frequency clustering results, "high-frequency interactive" content is identified, including important plot nodes and interactive selection points where users are concentrated; Use the time series model RIMA to predict changes in content popularity; Predict the future popularity of content by studying the interaction data trends over the past period of time; Recent interaction data is input into the time series model in real time to dynamically update the popularity prediction results so that the content priority reflects the current user interests.

5. The method for interactively displaying extensible content based on streaming media playback according to claim 4, characterized in that: The calculation formula of the clustering algorithm K-Means to classify the interactive behavior data and identify high-frequency interactive scenes is as follows: Initialize cluster centers Randomly select K initial cluster centers, calculate distances and assign data points to the nearest cluster center. For each interactive behavior data point x i , calculate its difference with all cluster centers μ j The Euclidean distance of: Where n is the number of feature dimensions, x ik is the data point x i The value of the K-th feature, μ jk is the cluster center μ j The value of the K-th feature; Assign all data points to the closest cluster to obtain clusters; then, update the μ of each cluster center j , calculate the mean of all data points belonging to this cluster: Among them, the C j represents the set of data points in cluster j, the |C j | is the number of data points in cluster j; Repeat steps 2 and 3 until the position of the cluster center no longer changes significantly and the preset upper limit of the number of iterations is reached; After clustering, analyze the feature center of each cluster; cluster centers with higher interaction frequency and dwell time can be identified as "high-frequency interaction scenes", and the central feature values ​​of these clusters represent the user's high interaction intention; define a threshold to determine whether a cluster belongs to a high-frequency interaction scene; for a cluster j, if the number of clicks and dwell time of its feature center are higher than the threshold, the cluster is defined as a high-frequency interaction scene cluster.

6. The method for interactively displaying extensible content based on streaming media playback according to claim 3, characterized in that: The S2-1 uses a dynamic partitioning algorithm to adjust the cache capacity based on priority scoring, places high-priority content in the CDN, combines the Least Recently Used and Least Frequently Used mixed replacement strategies, and uses the user-described clustering algorithm K-Means to group users, identify groups with different viewing behaviors, and distinguish users who prefer plot interaction from users who prefer scene exploration; according to the preferences of user groups, differentiated content priorities are assigned to different groups; the plot interaction preference group will prioritize caching core interactive content, while the exploration group will prioritize caching scene content.

7. The method for interactively displaying extensible content based on streaming media playback according to claim 6, characterized in that: The comprehensive priority score of the content is calculated using a weighted formula, which is as follows: Priority Rating = Interaction frequency+ heat Among them, the and The weights represent interaction frequency and content popularity respectively, and are adjusted dynamically according to business needs. Content is sorted according to priority scores. High-scoring content is marked as priority cache content, and low-scoring content is downgraded to non-priority cache content.

8. The method for interactively displaying extensible content based on streaming media playback according to claim 1, characterized in that: The S5 deep reinforcement learning algorithm predicts user interaction behavior including the following steps: Define the user's current viewing status, including current content type, selected interaction path, viewing time, and interaction frequency characteristics; Use a deep neural network to build a Q network model, input the user's current state characteristics, and output the Q value of each interactive operation, that is, the expected reward value of each action. The Q network formula is as follows: Q(s,a)=r+ Wherein, s represents the current state, a represents the current action, r represents the reward value, and stated Represent the next state and the optimal action respectively. ; The deep neural network is used to build a Q network model, the user's current state features are input, and the Q value of each interactive operation, that is, the expected reward value of each action, is output; the collected user historical behavior data is used to build a training set, the playback mechanism is used to improve the model stability, and the target network parameters are updated regularly; in an actual environment, the user's newly generated behavior data is collected in real time, and the collected behavior data is input into the model for fine-tuning.

9. The method for interactively displaying extensible content based on streaming media playback according to claim 8, characterized in that: Preloading predicted content using the HTTP / 2 ServerPush protocol includes the following steps: S5-1. Set a push list and use the high-probability content predicted by the Q network model as a push resource; once the server detects that the user starts loading a video clip, it immediately pushes the predicted content to the client according to the push list; S5-2. According to the predicted probability of the Q network model, different priorities are set for the pushed content; high-probability content is set to high priority and preloaded first; combined with the client cache strategy, ensure that the predicted content pushed is cached in the user device, and low-priority content is dynamically loaded after the user selects it; S5-3. Configure a loading control mechanism on the client to allow users to directly load the next predicted interactive content without delay when completing the current interaction; when the user's actual selection does not match the pushed content, the system will first extract the content from the cache; if the content does not exist in the cache, it will be loaded from the server immediately; S5-4. Update the reward mechanism of the reinforcement learning model by comparing the user's actual choice with the predicted result; regularly perform model training iterations and input the latest user behavior data into the Q network model.

10. The scalable content interactive display system based on streaming media playback is characterized by: The extensible content interactive display system based on streaming media playback includes rights 1-9 of the extensible content interactive display method based on streaming media playback, including a user behavior analysis module, a content management and distribution module, an interactive prediction and recommendation module, a preloading and cache management module, a network transmission and optimization module, a content display and rendering module, a data collection and feedback module and a security and authority management module.

Images