Acceleration service customization system based on user portrait

Abstract

The invention discloses a player emotion perception and experience optimization system based on artificial intelligence, and the system comprises a user portrait construction module which is used for generating a multi-dimensional user portrait vector; the scene prediction module is used for outputting a prediction result based on the user portrait vector and the historical behavior event sequence; the strategy generation module is used for dynamically generating a strategy vector and encoding the strategy vector into an acceleration service strategy object; the resource preparation module is used for registering the scheduling information to a service scheduling control table; the service monitoring module is used for collecting a service experience index set in real time, calculating a deviation vector and judging whether a strategy adjustment process is triggered or not; and the strategy optimization module is used for training the multi-scale residual convolutional neural network. According to the method, an active personalized acceleration service system is constructed by combining multi-dimensional user portrait modeling, adaptive strategy generation and a feedback optimization mechanism, so that the intelligent level of a network service system and the terminal user experience quality are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent service acceleration, and more particularly to a customized acceleration service system based on user profiles. Background Technology

[0002] In the context of the rapid development of internet services today, network acceleration and service performance optimization technologies have been widely applied in various scenarios such as video-on-demand, online games, remote work, cloud computing, and mobile applications. In order to improve user experience, service providers usually adopt traditional acceleration methods such as content delivery networks, edge computing node deployment, link optimization, and quality of service assurance mechanisms. These methods are mostly based on a unified network topology or application type for resource allocation and acceleration strategy configuration, and mainly focus on static or semi-static parameter adjustments such as data link optimization, bandwidth enhancement, and node selection.

[0003] Chinese invention patent with announcement number CN116155729B and announcement date of 2024.11.01 discloses a network acceleration method and related electronic equipment based on user traffic profile. This technology identifies the user's current data traffic status and avoids network acceleration when the user's data traffic is below the warning threshold, thereby effectively preventing the problem of additional charges due to excessive traffic usage.

[0004] However, with the surge in the number of end users and the increasing diversity of their behavior patterns, traditional acceleration mechanisms have gradually revealed problems such as poor adaptability, low resource utilization, and lagging policy response. Specifically, acceleration strategies lack personalization and cannot be customized in a targeted manner according to different users' usage behavior, terminal devices, geographical environment, and network status; policy response modes are mostly passive, that is, scheduling and resource preparation are only carried out after the user request arrives, resulting in increased latency for the first access; in addition, existing systems generally lack the ability to predict users' future usage behavior, cannot proactively perceive the changing trends of users' service needs, and cannot pre-allocate resources and configure policies in advance based on prediction results.

[0005] Therefore, there is an urgent need for a service optimization solution that can build and integrate behavioral prediction models based on multi-dimensional user profiles and has the ability to proactively accelerate strategy scheduling. In response to the above needs, this invention proposes a proactive acceleration service customization system and method based on user profiles, which overcomes the key problems in the prior art such as insufficient personalization, delayed response and lack of prediction feedback closed loop, and significantly improves user experience and system resource scheduling efficiency. Summary of the Invention

[0006] One objective of this invention is to propose an acceleration service customization system based on user profiles. This invention constructs a closed-loop acceleration service customization architecture consisting of a user profile building module, a scenario prediction module, a strategy generation module, a resource preparation module, a service monitoring module, and a strategy optimization module. It integrates multi-dimensional behavioral data collection and feature modeling mechanisms, employs a time-series prediction model with a multi-scale residual convolutional fusion attention mechanism, and achieves accurate prediction of future service scenarios for users. Furthermore, it combines reinforcement learning and a profile weight-driven strategy generation method to dynamically output personalized acceleration strategies, pre-completing content pre-fetching, edge resource configuration, and scheduling queue management. The service monitoring module collects service experience indicators in real time and feeds them back to the prediction model for adaptive optimization, enabling continuous evolution of acceleration strategies and proactive protection of service experience. This method effectively solves the problems of lack of personalization in acceleration strategies, delayed resource response, and disconnect between prediction and feedback mechanisms in existing technologies. It significantly improves the prediction accuracy, scheduling flexibility, and user satisfaction of the service system, and is applicable to intelligent acceleration service optimization in various scenarios such as content distribution, cloud computing, and mobile applications.

[0007] According to an embodiment of the present invention, a user profile-based acceleration service customization system includes: The user profile building module is used to collect basic user information, information on the terminal devices used by users, network environment information and behavioral data, extract key features, and generate multi-dimensional user profile vectors. The scene prediction module is used to construct time series samples based on user profile vectors and historical behavior event sequences, and then output prediction results through a multi-scale residual convolutional neural network. The strategy generation module is used to dynamically generate strategy vectors based on prediction results and user profile weight information, combined with the strategy template library, and encode them into acceleration service strategy objects. The resource preparation module is used to perform content scheduling, edge node computing resource allocation, scheduling queue configuration and lifecycle control before the prediction period specified by the strategy, and to register the scheduling information to the service scheduling control table. The service monitoring module is used to initiate the service delivery process, collect a set of service experience metrics in real time during user access, calculate the deviation vector, and determine whether to trigger the strategy adjustment process based on the current strategy execution effect. The strategy optimization module is used to train a multi-scale residual convolutional neural network based on gradient descent, with the bias vector as input.

[0008] Optional, The user's basic information includes the user's unique identifier, age, gender, account registration time, and location information; The terminal device information used by the user includes device type, operating system version, device model, screen resolution, processor model, memory capacity, and network connection type. The network environment information includes access method, network bandwidth, current network latency, packet loss rate, jitter value, and historical network status change trends; The behavioral data includes the application type accessed within different time periods, usage frequency, usage duration, operation event sequence, and event timestamp.

[0009] Optionally, the steps for constructing time-series samples based on user profile vectors and historical behavioral event sequences, and then outputting prediction results through a multi-scale residual convolutional neural network, are as follows: S21. Convert user profile vectors With historical behavioral event sequence The input sequence tensor is generated by splicing and merging the sequences. S22. Input the input sequence tensor into a multi-scale residual convolutional neural network, and concatenate the outputs of all branches to form a residual representation. ; S23. Residual representation Apply channel attention mechanism to calculate attention weight vector , of which The weights of each channel are: ; in, Indicates channel The characteristic of MLP is that it represents a multilayer perceptron network. Represents the natural exponential function; S24. Obtain the weighted feature matrix by weighting. The weighted feature matrix The input is fed into a bidirectional gated recurrent unit network for timing modeling to obtain the output vector sequence. Each output vector Corresponding prediction time point The characteristics of its use cases; S25, each Decoding into prediction results ,in, Indicates the target application type. This indicates the required service quality level, forming a set of predicted usage scenarios. , .

[0010] Optionally, the multi-scale residual convolutional neural network includes multiple parallel convolutional branches, each branch using a different kernel size to extract user behavior change features at different time scales.

[0011] Optionally, the steps of dynamically generating a strategy vector based on the prediction results and user profile weight information, combined with the strategy template library, and encoding it into an acceleration service strategy object are as follows: S31, Prediction Set Based on Usage Scenarios Each prediction result In The corresponding target application type is selected from a predefined acceleration strategy template library to find a matching basic strategy template. ; S32. Based on the feature dimension weights in the user profile vector, use a weighting function to personalize and optimize the strategy template, generating a strategy vector. : ; in, Represents the characteristic weighting matrix, The weighted mapping function outputs a policy vector. Includes personalized acceleration service configuration parameters for that predicted time point; S33, Regarding the strategy vector The parameters in the policy vector are verified, including resource availability detection, policy conflict investigation, and timing consistency verification, to ensure that the policy generation process meets the system resource scheduling constraints and behavior prediction consistency requirements. This includes content prefetching parameters, edge node resource allocation parameters, service scheduling priority parameters, and caching strategy parameters; S34. Transfer the validated policy vector Encoding as an acceleration service strategy object ,in, To predict the time point, For the target application type, To determine the required service quality level, a set of strategies is established. , .

[0012] Optionally, the steps of performing content scheduling, edge node computing resource allocation, scheduling queue configuration, and lifecycle control before the prediction time period specified by the strategy, and registering the scheduling information to the service scheduling control table are as follows: S41. Based on the acceleration service policy object Predicted time points With current system time The relationship between the calculation of the preparation time window , ,like ,in, If a preparation time threshold is set for the system, the resource preparation process will be triggered. S42, Based on the strategy vector The content prefetch parameters are used to perform content scheduling operations, transferring the predicted access content from the main content repository to the edge node cache closest to the user, forming a local cache set. , of which each Indicates a cached content block; S43, Based on the strategy vector The edge node resource allocation parameters in the edge computing node pool Select nodes that meet the load constraints Allocate the corresponding set of computing resources , of which each Including CPU cores, memory, and network ports; S44, Based on the strategy vector The service scheduling priority parameter in the target application type Create a scheduling queue And set priority levels The smaller the value, the higher the priority. S45, Based on the strategy vector The caching strategy parameters in the file apply to the cached content set. Configure lifecycle control strategies, including lifecycle duration. and replacement strategy , , ; S46. Register the edge node address, cache mapping table, scheduling queue information, and resource binding information to the service scheduling control table. And set the activation time as This is for subsequent user access to call.

[0013] Optionally, the steps for initiating the service delivery process, collecting a set of service experience metrics in real time during user access, calculating the deviation vector, and determining whether to trigger the strategy adjustment process based on the current strategy execution effect are as follows: S51, Arriving at the user's access time point When, enable the service scheduling control table. Record the acceleration resources and initiate the service delivery process; S52, In the user access application type During the process, a set of service experience metrics is collected in real time. The set of service experience metrics includes average access latency. Loading completion time lag rate and throughput ; S53. Set up the collected service experience metrics With the required service quality level Compare the target values ​​in the data and calculate the deviation vector. : ; in, This represents the difference between each actual indicator and the target value; S54. Based on the deviation vector Determine the effect of the current strategy execution; if any ,in, As an indicator If the tolerance threshold is exceeded, it is determined that the current service experience is abnormal, and the strategy adjustment process is triggered; otherwise, the feedback optimization process is initiated. Optionally, the operation steps of the strategy adjustment process are as follows: S61. Based on the current service session context, dynamically adjust cache scheduling parameters, edge node resource allocation, scheduling priority level, and resource invocation method to form a new strategy adjustment instruction. ; S62. Apply the adjusted parameters to the current service session execution cycle and update the service scheduling control table. It records the current adjustment behavior and corresponding indicator feedback, and stores them in the historical optimization record library for subsequent modeling and strategy training.

[0014] Optionally, the steps for training a multi-scale residual convolutional neural network using the bias vector as input and based on gradient descent are as follows: S71, the deviation vector Create training sample pairs and add them to the sample set. middle; S72, Using a sample set By minimizing the loss function Training a multi-scale residual convolutional neural network, loss function The calculation method is as follows: ; in, For the sample size, The number of metrics in the service experience metrics set. The difference between each actual indicator and the target value; S73. The trained multi-scale residual convolutional neural network is used as a new prediction model to update the prediction input in the policy generation module, thereby accelerating the adaptive evolution of the policy.

[0015] The beneficial effects of this invention are: 1. This invention constructs a data perception mechanism centered on user profiles, fully integrating users' basic information, device attributes, network environment, and usage behavior to generate high-dimensional user feature vectors, significantly improving the ability to identify user differences. On this basis, the system introduces a time series prediction model that integrates multi-scale residual convolution and attention mechanisms to accurately predict users' usage scenarios in specific future time periods, including usage time, application type, and service quality requirements. This provides a data foundation for subsequent resource scheduling and strategy formulation, realizing the transformation of services from "passive triggering" to "proactive planning".

[0016] 2. This invention dynamically generates customized acceleration service strategies by jointly modeling prediction results and user profiles. It completes preparatory work such as cache prefetching, edge computing resource pre-allocation, and scheduling queue configuration before user behavior occurs, avoiding the resource cold start and scheduling delay problems in the initial access phase of traditional solutions. This method effectively reduces initial loading time and request waiting latency, improving the instant response capability and resource utilization efficiency of the acceleration system.

[0017] 3. This invention collects user experience index data in real time during the service process, constructs feedback vectors, and uses them to update the parameters of the usage scenario prediction model. This enables adaptive optimization and continuous evolution of the acceleration strategy. This closed-loop mechanism not only improves the accuracy of the prediction model and the flexibility of system scheduling, but also solves the problem of lack of feedback closed loop and dynamic adjustment capabilities in traditional acceleration technologies. It significantly improves the intelligence level of the overall acceleration service system and the end-user's experience satisfaction. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a module structure diagram of the user profile-based acceleration service customization system proposed in this invention; Figure 2 The present invention provides an overall method flowchart for a user profile-based acceleration service customization system. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 and Figure 2 A user profile-based acceleration service customization system includes: The user profile building module is used to collect basic user information, information on the terminal devices used by users, network environment information and behavioral data, extract key features, and generate multi-dimensional user profile vectors. The scene prediction module is used to construct time series samples based on user profile vectors and historical behavior event sequences, and then output prediction results through a multi-scale residual convolutional neural network. The strategy generation module is used to dynamically generate strategy vectors based on prediction results and user profile weight information, combined with the strategy template library, and encode them into acceleration service strategy objects. The resource preparation module is used to perform content scheduling, edge node computing resource allocation, scheduling queue configuration and lifecycle control before the prediction period specified by the strategy, and to register the scheduling information to the service scheduling control table. The service monitoring module is used to initiate the service delivery process, collect a set of service experience metrics in real time during user access, calculate the deviation vector, and determine whether to trigger the strategy adjustment process based on the current strategy execution effect. The strategy optimization module is used to train a multi-scale residual convolutional neural network based on gradient descent, with the bias vector as input.

[0021] In this embodiment, The user's basic information includes the user's unique identifier, age, gender, account registration time, and location information; The terminal device information used by the user includes device type, operating system version, device model, screen resolution, processor model, memory capacity, and network connection type. The network environment information includes access method, network bandwidth, current network latency, packet loss rate, jitter value, and historical network status change trends; The behavioral data includes the application type accessed within different time periods, usage frequency, usage duration, operation event sequence, and event timestamp.

[0022] In this embodiment, the steps for constructing time-series samples based on user profile vectors and historical behavioral event sequences, and then outputting prediction results through a multi-scale residual convolutional neural network, are as follows: S21. Convert user profile vectors With historical behavioral event sequence The input sequence tensor is generated by splicing and merging the sequences. S22. Input the input sequence tensor into a multi-scale residual convolutional neural network, and concatenate the outputs of all branches to form a residual representation. ; S23. Residual representation Apply channel attention mechanism to calculate attention weight vector , of which The weights of each channel are: ; in, Indicates channel The characteristic of MLP is that it represents a multilayer perceptron network. Represents the natural exponential function; S24. Obtain the weighted feature matrix by weighting. The weighted feature matrix The input is fed into a bidirectional gated recurrent unit network for timing modeling to obtain the output vector sequence. Each output vector Corresponding prediction time point The characteristics of its use cases; S25, each Decoding into prediction results ,in, Indicates the target application type. This indicates the required service quality level, forming a set of predicted usage scenarios. , .

[0023] In this embodiment, the multi-scale residual convolutional neural network includes multiple parallel convolutional branches, each branch using a different kernel size to extract user behavior change features at different time scales.

[0024] In this embodiment, the steps of dynamically generating a strategy vector based on the prediction results and user profile weight information, combined with the strategy template library, and encoding it into an acceleration service strategy object are as follows: S31, Prediction Set Based on Usage Scenarios Each prediction result In The corresponding target application type is selected from a predefined acceleration strategy template library to find a matching basic strategy template. ; S32. Based on the feature dimension weights in the user profile vector, use a weighting function to personalize and optimize the strategy template, generating a strategy vector. : ; in, Represents the characteristic weighting matrix, The weighted mapping function outputs a policy vector. Includes personalized acceleration service configuration parameters for that predicted time point; S33, Regarding the strategy vector The parameters in the policy vector are verified, including resource availability detection, policy conflict investigation, and timing consistency verification, to ensure that the policy generation process meets the system resource scheduling constraints and behavior prediction consistency requirements. This includes content prefetching parameters, edge node resource allocation parameters, service scheduling priority parameters, and caching strategy parameters; S34. Transfer the validated policy vector Encoding as an acceleration service strategy object ,in, To predict the time point, For the target application type, To determine the required service quality level, a set of strategies is established. , .

[0025] In this embodiment, the steps of performing content scheduling, edge node computing resource allocation, scheduling queue configuration, and lifecycle control before the prediction time period specified by the strategy, and registering the scheduling information to the service scheduling control table are as follows: S41. Based on the acceleration service policy object Predicted time points With current system time The relationship between the calculation of the preparation time window , ,like ,in, If a preparation time threshold is set for the system, the resource preparation process will be triggered. S42, Based on the strategy vector The content prefetch parameters are used to perform content scheduling operations, transferring the predicted access content from the main content repository to the edge node cache closest to the user, forming a local cache set. , of which each Indicates a cached content block; S43, Based on the strategy vector The edge node resource allocation parameters in the edge computing node pool Select nodes that meet the load constraints Allocate the corresponding set of computing resources , of which each Including CPU cores, memory, and network ports; S44, Based on the strategy vector The service scheduling priority parameter in the target application type Create a scheduling queue And set priority levels The smaller the value, the higher the priority. S45, Based on the strategy vector The caching strategy parameters in the file apply to the cached content set. Configure lifecycle control strategies, including lifecycle duration. and replacement strategy , , ; S46. Register the edge node address, cache mapping table, scheduling queue information, and resource binding information to the service scheduling control table. And set the activation time as This is for subsequent user access to call.

[0026] In this embodiment, the steps for initiating the service delivery process, collecting a set of service experience metrics in real time during user access, calculating the deviation vector, and determining whether to trigger the strategy adjustment process based on the current strategy execution effect are as follows: S51, Arriving at the user's access time point When, enable the service scheduling control table. Record the acceleration resources and initiate the service delivery process; S52, In the user access application type During the process, a set of service experience metrics is collected in real time. The set of service experience metrics includes average access latency. Loading completion time lag rate and throughput ; S53. Set up the collected service experience metrics With the required service quality level Compare the target values ​​in the data and calculate the deviation vector. : ; in, This represents the difference between each actual indicator and the target value; S54. Based on the deviation vector Determine the effect of the current strategy execution; if any ,in, As an indicator If the tolerance threshold is exceeded, it is determined that the current service experience is abnormal, and the strategy adjustment process is triggered; otherwise, the feedback optimization process is initiated. In this embodiment, the operation steps of the strategy adjustment process are as follows: S61. Based on the current service session context, dynamically adjust cache scheduling parameters, edge node resource allocation, scheduling priority level, and resource invocation method to form a new strategy adjustment instruction. ; S62. Apply the adjusted parameters to the current service session execution cycle and update the service scheduling control table. It records the current adjustment behavior and corresponding indicator feedback, and stores them in the historical optimization record library for subsequent modeling and strategy training.

[0027] In this embodiment, the steps for training a multi-scale residual convolutional neural network using the bias vector as input and based on gradient descent are as follows: S71, the deviation vector Create training sample pairs and add them to the sample set. middle; S72, Using a sample set By minimizing the loss function Training a multi-scale residual convolutional neural network, loss function The calculation method is as follows: ; in, For the sample size, The number of metrics in the service experience metrics set. The difference between each actual indicator and the target value; S73. The trained multi-scale residual convolutional neural network is used as a new prediction model to update the prediction input in the policy generation module, thereby accelerating the adaptive evolution of the policy.

[0028] Example This embodiment uses the scenario of online video platform users watching high-definition videos on mobile devices during peak hours as an example to illustrate the application process and optimization effect of the "Proactive Acceleration Service Customization System Based on User Profile" proposed in this invention in a real business environment.

[0029] The user is a 20-year-old college student in Shenzhen, Guangdong. He frequently uses a 5G phone to watch 4K high-definition videos between 10:00 PM and 11:30 PM every night. Traditional network acceleration methods only perform CDN scheduling and node selection after the user initiates a video request. This results in a 4-6 second buffering wait during the initial video loading phase every night, and occasional image quality degradation or stuttering during playback, affecting the viewing experience. Especially during high-concurrency periods, the lack of advance awareness of user behavior and resource pre-allocation mechanisms leads to sudden increases in CDN node load, resulting in a significant decline in user experience.

[0030] After introducing the system of this invention, the system first collects the user's viewing habits, device usage characteristics, network environment status and content preferences over the past 30 days through the client SDK. The system constructs a multi-dimensional user profile and identifies the user as belonging to the "stable nighttime heavy HD video user" profile group. The user's terminal is a Xiaomi 12 Pro with a resolution of 3200×1440, an average available bandwidth of 38Mbps, and the usage frequency is concentrated after 22:00 every day. The user's content preference is high bitrate 4K movies and TV series.

[0031] Using the user profile information and behavioral sequence as input, the time series prediction model predicts that the user will continuously watch 4K video content from 22:00 to 23:30 for the next three consecutive days. The system then completes the following proactive preparations before 21:40 each day: caches the target video content in segments to the edge node in Longgang District, Shenzhen, which is closest to the user, and reserves node computing resources (including 1.5 CPU cores, 512MB cache, and 1Gbps egress bandwidth) as a guarantee; at the same time, it raises the user's scheduling policy priority to the highest local priority to ensure that the request enters the high-priority queue for execution as soon as it is issued.

[0032] When a user actually accesses the service, the service monitoring module records data of their entire viewing process. Statistical indicators include initial loading time, playback stuttering rate, number of bitrate switching times, and overall average latency. The system compares these experience indicators with the prediction results of the usage scenario and automatically adjusts the model prediction weights and scheduling parameters based on the actual deviation, forming an iterative update closed loop for the model.

[0033] After seven consecutive days of experimental observation, the service experience of the system of this invention and the traditional solution were compared horizontally under the same user and network environment. The results showed that after adopting the system of this invention, the user's first loading time was reduced from an average of 5.2 seconds to 1.1 seconds, the stuttering rate during playback was reduced from 6.4% to 0.7%, the video bitrate switching frequency was reduced from an average of 4.1 times per day to 0.3 times, and the overall latency was reduced from 280ms to 76ms, which significantly improved the smoothness of the user experience and the stability of the platform service quality.

[0034] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A user profile-based acceleration service customization system, characterized in that, include: The user profile building module is used to collect basic user information, information on the terminal devices used by users, network environment information and behavioral data, extract key features, and generate multi-dimensional user profile vectors. The scene prediction module is used to construct time series samples based on user profile vectors and historical behavior event sequences, and then output prediction results through a multi-scale residual convolutional neural network. The strategy generation module is used to dynamically generate strategy vectors based on prediction results and user profile weight information, combined with the strategy template library, and encode them into acceleration service strategy objects. The resource preparation module is used to perform content scheduling, edge node computing resource allocation, scheduling queue configuration and lifecycle control before the prediction period specified by the strategy, and to register the scheduling information to the service scheduling control table. The service monitoring module is used to initiate the service delivery process, collect a set of service experience metrics in real time during user access, calculate the deviation vector, and determine whether to trigger the strategy adjustment process based on the current strategy execution effect. The strategy optimization module is used to train a multi-scale residual convolutional neural network based on gradient descent, with the bias vector as input.

2. The user profile-based acceleration service customization system according to claim 1, characterized in that, The user's basic information includes the user's unique identifier, age, gender, account registration time, and location information; The terminal device information used by the user includes device type, operating system version, device model, screen resolution, processor model, memory capacity, and network connection type. The network environment information includes access method, network bandwidth, current network latency, packet loss rate, jitter value, and historical network status change trends; The behavioral data includes the application type accessed within different time periods, usage frequency, usage duration, operation event sequence, and event timestamp.

3. The user profile-based acceleration service customization system according to claim 1, characterized in that, The steps for constructing time-series samples based on user profile vectors and historical behavioral event sequences, and then outputting prediction results through a multi-scale residual convolutional neural network, are as follows: S21. Concatenate and fuse the user profile vector with the historical behavior event sequence to generate the input sequence tensor; S22. Input the input sequence tensor into a multi-scale residual convolutional neural network, and concatenate the outputs of all branches to form a residual representation. ; S23. Residual representation Apply channel attention mechanism to calculate attention weight vector ; S24. Obtain the weighted feature matrix by weighting. The weighted feature matrix The input is fed into a bidirectional gated recurrent unit network for timing modeling to obtain the output vector sequence. Each output vector Corresponding prediction time point The characteristics of its use cases; S25. Convert each output vector Decoding into prediction results ,in, Indicates the target application type. This indicates the required service quality level, forming a set of predicted usage scenarios. , .

4. The user profile-based accelerated service customization system according to claim 3, wherein the multi-scale residual convolutional neural network comprises multiple parallel convolutional branches, each branch employing a different convolutional kernel size.

5. The user profile-based acceleration service customization system according to claim 3, characterized in that, The steps for dynamically generating policy vectors based on prediction results and user profile weight information, combined with the policy template library, and encoding them into acceleration service policy objects are as follows: S31, Prediction Set Based on Usage Scenarios Each prediction result In The corresponding target application type is selected from a predefined acceleration strategy template library to find a matching basic strategy template. ; S32. Based on the feature dimension weights in the user profile vector, use a weighting function to personalize and optimize the strategy template, generating a strategy vector. ; S33, Regarding the strategy vector The parameters in the policy vector are verified, including resource availability detection, policy conflict investigation, and timing consistency verification. This includes content prefetching parameters, edge node resource allocation parameters, service scheduling priority parameters, and caching strategy parameters; S34. Transfer the validated policy vector Encoding as an acceleration service strategy object ,in, To predict the time point, For the target application type, To determine the required service quality level, a set of strategies is established. , .

6. The user profile-based acceleration service customization system according to claim 5, characterized in that, The steps for performing content scheduling, edge node computing resource allocation, scheduling queue configuration, and lifecycle control before the prediction time period specified by the strategy, and registering the scheduling information to the service scheduling control table are as follows: S41. Based on the acceleration service policy object Predicted time points With current system time The relationship between the calculation of the preparation time window ,like ,in, If a preparation time threshold is set for the system, the resource preparation process will be triggered. S42, Based on the strategy vector The content prefetch parameters are used to perform content scheduling operations, transferring the predicted access content from the main content library to the edge node cache closest to the user, forming a local cache set; S43, Based on the strategy vector The edge node resource allocation parameters are used to select nodes that meet the load constraints from the edge computing node pool and allocate the corresponding set of computing resources. , of which each Including CPU cores, memory, and network ports; S44, Based on the strategy vector The service scheduling priority parameter in the target application type Create a scheduling queue And set priority levels ; S45, Based on the strategy vector The caching strategy parameters in the file apply to the cached content set. Configure lifecycle control strategies, including lifecycle duration. and replacement strategy ; S46. Register the edge node address, cache mapping table, scheduling queue information, and resource binding information to the service scheduling control table. And set the activation time as .

7. The user profile-based acceleration service customization system according to claim 6, characterized in that, The steps for initiating the service delivery process, collecting service experience metrics in real time during user access, calculating the deviation vector, and determining whether to trigger a policy adjustment process based on the current policy execution effect are as follows: S51, Arriving at the user's access time point When, enable the service scheduling control table. Record the acceleration resources and initiate the service delivery process; S52, In the user access application type During the process, a set of service experience metrics is collected in real time. The set of service experience metrics includes average access latency, loading completion time, lag rate, and throughput. S53. Set up the collected service experience metrics With the required service quality level Compare the target values ​​in the data and calculate the deviation vector. ; S54. Based on the deviation vector Determine the effect of the current strategy execution; if any ,in, As an indicator If the tolerance threshold is reached, it is determined that the current service experience is abnormal, triggering the policy adjustment process; otherwise, it enters the feedback optimization process.

8. The user profile-based acceleration service customization system according to claim 7, characterized in that, The operational steps of the strategy adjustment process are as follows: S61. Based on the current service session context, dynamically adjust cache scheduling parameters, edge node resource allocation, scheduling priority level, and resource invocation method to form a new strategy adjustment instruction. ; S62. Apply the adjusted parameters to the current service session execution cycle and update the service scheduling control table. And record the current adjustment behavior and the corresponding indicator feedback.

9. The user profile-based acceleration service customization system according to claim 7, characterized in that, The steps for training a multi-scale residual convolutional neural network using the bias vector as input and gradient descent are as follows: S71, the deviation vector Create training sample pairs and add them to the sample set. middle; S72, Using a sample set By minimizing the loss function Training a multi-scale residual convolutional neural network; S73. Use the trained multi-scale residual convolutional neural network as a new prediction model.

Images