Network service method, apparatus, and medium

Abstract

The present disclosure provides a network service method, device and medium, relating to the technical field of network, the method comprises: receiving a network service request of a user; predicting a plurality of predicted network service tasks triggered by subsequent user behaviors according to the network service request and a user behavior prediction model; and allocating a computing service node for the network service request according to the plurality of predicted network service tasks and a dynamic load balancing strategy, so that the computing service node prepares to call network service resources for the network service request according to the plurality of predicted network service tasks. The present disclosure predicts user behaviors, obtains subsequent network service tasks triggered by the network service request, combines the mode of dynamic load balancing and preparation of resource calling, and allocates computing nodes and calls network service resources for the predicted network service tasks in advance, so as to respond to user demand more timely and improve the network use experience of the user.

Description

Technical Field

[0001] This disclosure relates at least to the field of network technology, and in particular to a method for providing network services.

[0002] Network service devices and computer-readable storage media. Background Technology

[0003] In handling user network service requests, existing technologies still face some challenges in practical applications. For example, a sudden surge in data flow may cause system response delays, affecting user experience. To improve response speed, strategies such as dynamic load balancing may be used in some cases, but without prior prediction and preparation, relying solely on dynamic load balancing is insufficient to fully meet user needs. Summary of the Invention

[0004] The technical problem to be solved by this disclosure is to provide a network service method, network service apparatus, and computer-readable storage medium to address the above-mentioned shortcomings, so as to solve the problem of how to respond promptly to users' continuous network service needs through prediction and preparation.

[0005] In a first aspect, this disclosure provides a network service method, the method comprising:

[0006] Receive user network service requests;

[0007] Based on network service requests and user behavior prediction models, predict that subsequent user behavior will trigger several predictive network service tasks to be executed.

[0008] Based on several predicted network service tasks and dynamic load balancing strategies, computing service nodes are allocated to network service requests, so that the computing service nodes can prepare to call network service resources for network service requests according to several predicted network service tasks.

[0009] Furthermore, among which:

[0010] Network service requests include requests from users to browse products on e-commerce platforms;

[0011] Several predictive network service tasks include browsing products, switching browsing modes, adding items to the cart, and completing a purchase.

[0012] Furthermore, based on the network service request and user behavior prediction model, several predictive network service tasks that will be triggered by subsequent user behavior are predicted, specifically including:

[0013] The first historical user behavior data is obtained based on the network service request. The first historical user behavior data includes several historical network service tasks that were executed after the user made the same type of network service request in the past, as well as the historical execution waiting time and historical execution probability of each historical network service task.

[0014] Using a user behavior prediction model based on first-historical user behavior data, predict several predictive network service tasks that will be executed due to subsequent user behavior, as well as the predicted execution waiting time and predicted execution probability of each predictive network service task.

[0015] Furthermore, before using the user behavior prediction model, the method also includes:

[0016] The user behavior prediction model needs to be updated according to the following formula: F = a*E + b*G, where a and b are balancing factors, E is the model prediction error obtained from the second historical user behavior data, and G represents the model complexity.

[0017] In response to the need to update the user behavior prediction model for evaluation, the user behavior prediction model is updated based on third historical user behavior data, which includes second historical user behavior data and is longer than first historical user behavior data.

[0018] Furthermore, based on several predicted network service tasks and dynamic load balancing strategies, computing service nodes are allocated to network service requests, specifically including:

[0019] Establish a real-time streaming task processing queue for network service requests, and add several predicted network service tasks to the real-time streaming task processing queue based on the predicted execution waiting time and the predicted execution probability.

[0020] Based on the load value and number of load tasks of the computing service node, allocate one or more computing service nodes to the real-time streaming task processing queue.

[0021] Furthermore, based on the predicted execution wait time and predicted execution probability, several predicted network service tasks are added to the real-time streaming task processing queue, specifically including:

[0022] Among a number of prediction network service tasks, the first prediction network service task whose prediction execution probability is greater than a first threshold is obtained.

[0023] Obtain the required network service resources when each first prediction network service task is executed;

[0024] Based on the required network service resource quantity U and the predicted execution waiting time T, obtain the resource preparation call priority of each first predicted network service task: P = k1*U + k2*T, where k1 and k2 are adjustment coefficients with opposite positive and negative values;

[0025] According to the priority of resource preparation and invocation, each first prediction network service task is added to the real-time stream task processing queue.

[0026] Furthermore, based on the load value and number of tasks of the computing service nodes, one or more computing service nodes are allocated to the real-time streaming task processing queue, specifically including:

[0027] Obtain the prediction execution time of each first prediction network service task in the real-time streaming task processing queue;

[0028] Obtain the predicted load value R and the number of predicted load tasks N for each computing service node at each predicted execution time;

[0029] The computational service node with the smallest value of W = α*R + β*N is selected and assigned to the corresponding first prediction network service task, where α and β are the weight coefficients of R and N, respectively.

[0030] Furthermore, the computing service node prepares to invoke network service resources for network service requests based on several predicted network service tasks, specifically including:

[0031] Obtain the pre-execution time of each first prediction network service task before each prediction execution time, in order to call network service resources;

[0032] The computing service node obtains the stored data that each first prediction network service task needs to access;

[0033] The computing service nodes will call the required stored data to the hot data layer during their respective pre-execution times.

[0034] Secondly, this disclosure provides a network service apparatus, the apparatus comprising:

[0035] The receiving module is used to receive network service requests from users.

[0036] The prediction module, connected to the receiving module, is used to predict several predictive network service tasks that will be triggered by subsequent user behavior based on network service requests and user behavior prediction models.

[0037] The allocation module, connected to the prediction module, is used to allocate computing service nodes to network service requests based on several predicted network service tasks and dynamic load balancing strategies, so that the computing service nodes can prepare to call network service resources for network service requests based on several predicted network service tasks.

[0038] Thirdly, this disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the network service method described above.

[0039] This disclosure provides a network service method, network service device, and computer-readable storage medium. By predicting user behavior and obtaining subsequent network service tasks that will be triggered by network service requests, and combining dynamic load balancing and pre-calling resource modes, computing nodes and network service resources are allocated in advance for the predicted network service tasks, which can respond to user needs more promptly and improve the user's network experience. Attached Figure Description

[0040] Figure 1 This is a flowchart of a network service method according to an embodiment of this disclosure;

[0041] Figure 2 This is a schematic diagram of the structure of a network service device according to an embodiment of the present disclosure;

[0042] Figure 3 This is a diagram showing the key steps of another network service method according to an embodiment of this disclosure;

[0043] Figure 4 This is a flowchart of a dynamic load balancing method for network services according to an embodiment of this disclosure;

[0044] Figure 5 This is a flowchart of another dynamic load balancing method for network services according to an embodiment of this disclosure;

[0045] Figure 6 This is a flowchart of a network service data standardization processing method according to an embodiment of this disclosure;

[0046] Figure 7 This is a flowchart of a method combining dynamic load balancing and real-time streaming technology for network services according to an embodiment of this disclosure;

[0047] Figure 8 This is a flowchart of a real-time streaming technology method for network services according to an embodiment of this disclosure;

[0048] Figure 9 This is a flowchart of a method for predicting user behavior in a network service according to an embodiment of this disclosure;

[0049] Figure 10 This is a flowchart of a network service data hierarchical storage method according to an embodiment of this disclosure;

[0050] Figure 11 This is a flowchart of another dynamic load balancing method for network services according to an embodiment of this disclosure;

[0051] Figure 12 This is a flowchart of another dynamic load balancing method for network services according to an embodiment of this disclosure;

[0052] Figure 13This is a flowchart of a network service data storage optimization method according to an embodiment of this disclosure;

[0053] Figure 14 This is a flowchart of a network service data transmission method according to an embodiment of this disclosure;

[0054] Figure 15 This is a flowchart of a network service data archiving method according to an embodiment of this disclosure. Detailed Implementation

[0055] To enable those skilled in the art to better understand the technical solutions of this disclosure, the embodiments of this disclosure will be further described in detail below with reference to the accompanying drawings.

[0056] It is understood that the specific embodiments and accompanying drawings described herein are for illustrative purposes only and are not intended to limit the scope of this disclosure.

[0057] It is understood that, without conflict, the various embodiments and features in the embodiments of this disclosure can be combined with each other.

[0058] It is understood that, for ease of description, only the parts relevant to this disclosure are shown in the accompanying drawings, while parts unrelated to this disclosure are not shown in the drawings.

[0059] It is understood that each module or unit involved in the embodiments of this disclosure may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple modules or units may be integrated into one entity structure.

[0060] It is understood that, without conflict, the functions and steps marked in the flowcharts and block diagrams of this disclosure may occur in a different order than that marked in the accompanying drawings.

[0061] It is understood that the flowcharts and block diagrams of this disclosure illustrate the architecture, functions, and operations of possible implementations of systems, apparatuses, devices, and methods according to various embodiments of this disclosure. Each block in a flowchart or block diagram may represent a module, unit, program segment, or code, containing executable instructions for implementing the specified function. Furthermore, each block or combination of blocks in the block diagrams and flowcharts may be implemented using hardware-based devices to implement the specified function, or using a combination of hardware and computer instructions.

[0062] It is understood that the modules and units involved in the embodiments of this disclosure can be implemented by software or by hardware, for example, the modules and units can be located in a processor.

[0063] Example 1:

[0064] like Figure 1As shown, this disclosure provides a network service method, the method comprising:

[0065] S1. Receive user's network service request;

[0066] S2. Based on the network service request and user behavior prediction model, predict the several predictive network service tasks that will be executed when the user behavior occurs.

[0067] S3. Based on several predicted network service tasks and dynamic load balancing strategies, allocate computing service nodes to network service requests so that the computing service nodes can prepare to call network service resources for network service requests based on several predicted network service tasks.

[0068] In this embodiment, by predicting user behavior and obtaining the subsequent network service tasks triggered by network service requests, and combining dynamic load balancing and pre-allocating resource requests, computing nodes and network service resources are allocated in advance for the predicted network service tasks. This allows for more timely responses to user needs and improves the user's network experience. Figure 2 As shown, the apparatus corresponding to the method includes: a receiving module 1 for executing step S1, a prediction module 2 for executing step S2, and an allocation module 3 for executing step S3. This embodiment implements a predictive dynamic load balancing method. Compared with existing dynamic load balancing strategies, it can pre-select suitable computing service nodes by predicting user behavior, and the computing service nodes can also prepare resources for predicting network service tasks.

[0069] Specifically, this embodiment presents a method for proactively perceiving and improving user satisfaction in network services based on big data and AI (Artificial Intelligence). Through big data analytics and AI algorithms, it enables real-time monitoring of user behavior and feedback, predicts and identifies potential problems in advance, and takes corresponding measures to improve user satisfaction and service quality. This addresses the issue of system response delays and negatively impacting user experience caused by sudden increases in data flow during high-concurrency access scenarios. It also considers addressing the problem of data parsing errors often caused by inconsistent data formats, which increases the complexity of data processing, and achieves the integration of multi-source heterogeneous data.

[0070] More specifically, the method of this embodiment is as follows: Figure 3 As shown, this method addresses typical problems in network services through five key steps, including system response latency under high concurrency access, data parsing errors under multi-source heterogeneous data integration, data processing latency under real-time analysis requirements, prediction accuracy decline under dynamic changes in user behavior, and cost optimization in large-scale data storage. Figure 3The five steps listed are key technical features of this embodiment. They do not represent a specific execution order, nor are they indispensable. Rather, they are optional steps that can be selected based on the specific problem. Each step and how it solves these problems are described in detail below.

[0071] S101 intelligently distributes high-concurrency access requests through a dynamic load balancing strategy to reduce system response latency caused by sudden surges in data flow. This means that when a large number of users access network services simultaneously, the system can automatically identify and balance the load of different servers, rationally distributing requests to the most suitable server, thereby preventing response delays caused by excessive load on any particular server. Specifically, when an e-commerce platform faces concentrated access from millions of users during a shopping festival, the intelligent load balancing algorithm dynamically distributes requests to ensure that each user's access request receives a timely response, thus improving the overall user experience. For example, when the system detects that a server's load is approaching saturation, it automatically routes new requests to other less loaded servers, ensuring that all requests can be processed quickly.

[0072] S102, based on the data format conversion and standardization module, automatically parses and unifies the format of multi-source heterogeneous data to avoid data parsing errors. The core function of this module is to automatically process data from different sources and formats, converting it into a unified standard format to avoid parsing errors caused by inconsistent data formats. For example, in a smart city project, it is necessary to integrate data from various sensors and devices, such as traffic monitoring cameras, environmental monitoring stations, and public safety equipment, whose data formats differ. The data format conversion and standardization module can automatically parse and unify this data into a standard format, facilitating subsequent data processing and analysis and ensuring data consistency and accuracy.

[0073] S103 utilizes real-time stream processing technology to accelerate data processing workflows, meeting real-time analysis needs and avoiding delays in analysis results. Real-time stream processing technology can process data simultaneously with its generation, avoiding the latency issues of traditional batch processing methods. Specifically, in the financial sector, the real-time requirements of trading systems are extremely high; even a millisecond delay can cause significant economic losses. By introducing real-time stream processing technology, trading systems can immediately analyze the latest market data and generate trading decisions in real time. For example, when the stock market experiences sudden and significant fluctuations, the real-time stream processing engine can quickly capture this information and trigger preset trading strategies, ensuring that the trading platform's decisions and executions are completed in the shortest possible time, thereby minimizing losses.

[0074] S104. The user behavior prediction model is periodically updated through a continuous online learning mechanism to improve prediction accuracy in the face of dynamically changing user behavior. User habits frequently change, and a single static model often struggles to adapt to these changes. Therefore, by employing an online learning mechanism, the system can periodically train and update the model based on the latest user behavior data, thereby continuously improving the model's prediction accuracy. For example, on online video platforms, users' viewing preferences change with time and content. Through the online learning mechanism, the platform can continuously adjust its recommendation algorithm based on users' recent viewing history and interaction behavior, making recommended content more aligned with users' current interests and improving user satisfaction. Specifically, if a user has recently been watching science fiction movies frequently, the model will gradually adjust, increasing the recommendation weight for this type of content.

[0075] S105 employs a tiered storage strategy for efficient management of large-scale data, optimizing storage costs and resource utilization. This strategy optimizes storage costs and resource utilization by categorizing and storing different types of data. Specifically, in enterprise data warehouses, hot data (frequently accessed data) is typically stored on high-performance but expensive storage media, such as solid-state drives (SSDs); cold data (infrequently accessed historical data) is stored on low-cost storage media, such as tape or object storage. In this way, enterprises can significantly reduce storage costs while maintaining access performance. For example, telecommunications companies typically need to store large amounts of call records and traffic logs, but the access frequency of this data varies greatly. Tiered storage allows active user data to be stored on high-performance SSDs, while inactive older data is archived in low-cost object storage, ensuring fast access to hot data while effectively controlling overall storage costs.

[0076] Through the five steps described above, the method for proactively perceiving and improving satisfaction in network services based on big data and AI effectively solves problems such as high-concurrency response latency, multi-source heterogeneous data integration errors, real-time data analysis latency, decreased accuracy of user behavior prediction, and large-scale data storage costs in network services, providing users with a stable, efficient, and intelligent network service experience.

[0077] against Figure 3 The method shown in this embodiment can provide at least two examples of each step, one example being as follows: Figure 4-10 As shown, another example is... Figure 11-15 As shown. Below, we will first combine... Figure 4-10 The method of this embodiment is explained.

[0078] In one embodiment, wherein:

[0079] Network service requests include requests from users to browse products on e-commerce platforms;

[0080] Several predictive network service tasks include browsing products, switching browsing modes, adding items to the cart, and completing a purchase.

[0081] In this embodiment, the method is illustrated with a specific application scenario. When users shop on e-commerce platforms, they typically exhibit both common network behavior characteristics, such as adding items to their cart and completing purchases after browsing similar products multiple times, and personalized network behavior characteristics, such as the number of products viewed, the duration of browsing a single item, and the timing of the user's purchase decision. The method in this embodiment analyzes the user's past shopping behavior characteristics to predict the subsequent actions and timing that the user's current product browsing will trigger. The network services required for these actions are pre-prepared as predicted network service tasks and added to the real-time streaming task processing queue for the user's corresponding network service requests. Corresponding service data is prepared in advance for these tasks, allowing the system to respond quickly when the user actually triggers these tasks, thereby improving the user experience. It is understood that the application of this method is not limited to this scenario; in other scenarios, the same approach of analyzing the user's historical behavior characteristics to predict future network behavior characteristics based on current network behavior and preparing data in advance is also possible.

[0082] In one implementation, based on network service requests and a user behavior prediction model, several predictive network service tasks that will be executed due to subsequent user behavior are predicted, specifically including:

[0083] The first historical user behavior data is obtained based on the network service request. The first historical user behavior data includes several historical network service tasks that were executed after the user made the same type of network service request in the past, as well as the historical execution waiting time and historical execution probability of each historical network service task.

[0084] Using a user behavior prediction model based on first-historical user behavior data, predict several predictive network service tasks that will be executed due to subsequent user behavior, as well as the predicted execution waiting time and predicted execution probability of each predictive network service task.

[0085] In this embodiment, one of the key aspects of predicting user behavior is predicting the actions the user will perform and when those actions will occur. Specifically, among a series of predictive network tasks the user may perform, it's necessary to predict the probability of each task being executed and the duration of each task's execution. Taking online shopping as an example, it might be necessary to predict how many similar products the user will browse before making a purchase decision, and how long each product will be browsed. Regarding the selection of historical data for prediction, for example, during a shopping festival, data from previous shopping festivals might be used, combined with data from the user's pre-sale period during the current shopping festival. These can be set based on actual conditions and experience. The key is to obtain the prediction execution waiting time and prediction execution probability for each predictive network service task.

[0086] In one embodiment, before using the user behavior prediction model, the method further includes:

[0087] The user behavior prediction model needs to be updated according to the following formula: F = a*E + b*G, where a and b are balancing factors, E is the model prediction error obtained from the second historical user behavior data, and G represents the model complexity.

[0088] In response to the need to update the user behavior prediction model for evaluation, the user behavior prediction model is updated based on third historical user behavior data, which includes second historical user behavior data and is longer than first historical user behavior data.

[0089] In this embodiment, in order to make the user behavior prediction model meet the usage requirements, the model is updated and trained. The decision is made by comprehensively evaluating the cost and benefits. The model prediction error is calculated based on the latest historical user behavior data, and the model complexity is combined to evaluate whether the model needs to be updated. If an update is needed, historical data over a longer period of time is collected for update training.

[0090] For example, such as Figure 9 As shown, a continuous online learning mechanism is used to periodically update the user behavior prediction model. Specifically, this mechanism includes:

[0091] S701 collects user interaction data in real time, including user clicks, browsing, and search history. This step is to obtain the latest user behavior information, thereby ensuring that the model can capture the user's latest preferences and behavioral patterns. For example, on an e-commerce platform, the system records every action a user takes, such as browsing products, adding items to their cart, and completing a purchase.

[0092] S702 utilizes this data to perform online incremental learning on existing user behavior models. This is achieved by continuously updating the model's parameters, enabling the model to adapt to the latest user behavior patterns. Specifically, the model is fine-tuned each time new user behavior data is received, gradually optimizing its predictive capabilities. This incremental learning method is more efficient than offline retraining and can reflect changes in user behavior more quickly.

[0093] S703: Periodically evaluate the model's predictive performance and adjust model parameters based on the evaluation results. Evaluations typically involve comparing the model's predictions with actual user behavior to measure the model's accuracy and effectiveness. If the evaluation results show a decline in model performance or room for improvement, the model parameters will be adjusted accordingly. For example, in a periodic evaluation, if the model's accuracy in predicting user purchase probabilities decreases, the system will add more historical purchase data from more users to improve the model.

[0094] S704, calculate whether the model needs to be updated based on the prediction error E and model complexity G, using the formula F = a*E + b*G, where a and b are balancing factors. Here, the prediction error E represents the difference between the model's predicted value and the actual value, ranging from 0 to positive infinity; a smaller value indicates a more accurate prediction. Model complexity G reflects the structural complexity of the model, also ranging from 0 to positive infinity; a larger value indicates a more complex model. Balancing factors a and b are used to adjust the weight of E and G in the overall evaluation function F, and their optimal values ​​are usually determined through multiple experiments. The meaning of formula F is that the larger the overall evaluation function F, the more urgent the need for model updates. This setting ensures that while considering prediction accuracy, it also avoids overfitting caused by an overly complex model. Specifically, in a certain embodiment, if F exceeds a preset threshold (e.g., 1.5), the model update process is triggered; otherwise, the existing model remains unchanged.

[0095] In one embodiment, computing service nodes are allocated to network service requests based on several predicted network service tasks and dynamic load balancing strategies, specifically including:

[0096] Establish a real-time streaming task processing queue for network service requests, and add several predicted network service tasks to the real-time streaming task processing queue based on the predicted execution waiting time and the predicted execution probability.

[0097] Based on the load value and number of load tasks of the computing service node, allocate one or more computing service nodes to the real-time streaming task processing queue.

[0098] In this embodiment, based on the prediction of user network behavior, the pre-processing of the predicted network service task is achieved by combining real-time stream queuing technology and dynamic network load balancing strategy.

[0099] For example, such as Figure 7 As shown, dynamic load balancing and real-time stream processing technologies accelerate the data processing flow as follows:

[0100] S501. Data is immediately added to the processing queue upon arrival, employing a distributed stream processing architecture to enhance parallel processing capabilities. Specifically, when data is generated, the system immediately adds it to the processing queue, utilizing a distributed computing framework to distribute the data across multiple processing nodes, thereby achieving parallel processing and improving overall processing speed. This step ensures the efficiency and real-time nature of data processing. For example, in an online customer service system, whenever a user submits a question or request, the system immediately adds it to the processing queue, processing it in parallel across multiple servers to quickly respond to user needs and improve user satisfaction.

[0101] S502. To further improve processing efficiency, this technology utilizes in-memory computing and caching to reduce I / O (Input, Output) operation time. By caching frequently used data in memory, frequent disk reads and writes are avoided, significantly reducing data access latency. In a network monitoring system based on big data analytics, for example, the system might cache common monitoring metrics in memory to generate reports or alerts in real time, eliminating the need to read data from the disk each time, thereby improving system response speed.

[0102] S503. Based on the results of real-time data analysis, the system dynamically adjusts the strategy for processing subsequent data streams, such as prioritizing critical data streams. Specifically, through real-time analysis of incoming data, data streams that have a significant impact on service quality are identified, and resources are allocated to them for processing in a prioritized manner. For example, in a recommendation system of an e-commerce platform, the system will prioritize processing the click streams of active users based on the real-time analysis results of user behavior data, and promptly push personalized product recommendations to improve user experience.

[0103] S504. Determine if the performance metric P of the processing node is greater than the threshold θ, using the formula P > θ, where P represents the current performance of the processing node, usually expressed as a percentage, ranging from 0% to 100%; θ is a preset performance threshold, for example, set to 80%. If it exceeds the threshold, the data flow routing strategy is optimized to improve processing speed. Specifically, when the performance of a processing node exceeds the preset threshold, the system automatically adjusts the data flow routing, distributing more tasks to other idle or higher-performing nodes, thereby ensuring the efficient operation of the entire system. For example, in a video streaming system, when the CPU utilization of a server consistently exceeds 80%, the system automatically redirects some video streams to servers with lower loads, ensuring the stability and smoothness of video transmission. The purpose of this setting formula is to avoid a single node becoming a performance bottleneck, ensuring the efficient operation of the entire system through dynamic optimization of resource allocation. Setting the threshold to 80% is to balance resource utilization and system stability, because an excessively high threshold may lead to wasted system resources, while an excessively low threshold may easily cause frequent resource adjustments, increasing system overhead.

[0104] In one embodiment, based on the predicted execution waiting time and the predicted execution probability, several predicted network service tasks are added to the real-time streaming task processing queue, specifically including:

[0105] Among a number of prediction network service tasks, the first prediction network service task whose prediction execution probability is greater than a first threshold is obtained.

[0106] Obtain the required network service resources when each first prediction network service task is executed;

[0107] Based on the required network service resource quantity U and the predicted execution waiting time T, obtain the resource preparation call priority of each first predicted network service task: P = k1*U + k2*T, where k1 and k2 are adjustment coefficients with opposite positive and negative values;

[0108] According to the priority of resource preparation and invocation, each first prediction network service task is added to the real-time stream task processing queue.

[0109] In this embodiment, the real-time streaming task processing queue technology is first used to prepare network service resources for invocation. By evaluating the resource invocation priority of each task, tasks are added to the queue sequentially. This priority considers the remaining waiting time for the task and the amount of service data that the task needs to download for the user, thereby determining their pre-execution order. Of course, it is also possible to predict the actual execution order of network service tasks by simply rearranging the tasks according to time sequence.

[0110] For example, such as Figure 8As shown, real-time stream processing technology is used to accelerate the data processing flow. The specific steps are as follows:

[0111] S601 involves real-time monitoring of the data flow to detect any anomalies or bottlenecks during processing. The core purpose of this step is to ensure the stability and efficiency of data processing. Real-time monitoring of the data flow allows for the timely identification and resolution of potential problems, preventing service interruptions or performance degradation due to data backlog or system failures. For example, in a network service, if a surge in data traffic is detected at a processing node in real-time, it may indicate that the node is experiencing abnormal access requests. The system will quickly focus on monitoring that node and take measures such as increasing resource allocation or redirecting traffic to restore its normal operation.

[0112] S602 collects system resource utilization data in real time for dynamic resource allocation adjustments. This measure enables the system to flexibly allocate resources according to actual needs, avoiding resource waste and overload. By continuously collecting data on CPU (Central Processing Unit), memory, and network bandwidth usage, the system's load status can be accurately determined, thereby achieving optimal resource allocation. Specifically, in a big data analytics platform, when a computing node's CPU utilization remains high for an extended period, the system automatically allocates resources from some low-load nodes to supplement it, thereby balancing the workload of each node and improving overall processing capacity.

[0113] S603 utilizes a real-time data feedback loop to optimize processing logic, thereby improving overall processing efficiency. By analyzing real-time data, it continuously improves and optimizes the algorithm to better adapt to current workload patterns. This process not only enhances the system's self-learning and adaptive capabilities but also significantly reduces latency and improves user experience. For example, in an AI-based content recommendation system, the recommendation strategy is dynamically adjusted based on real-time user interaction data, making recommended content more aligned with user interests and increasing click-through rates and user retention.

[0114] S604 calculates task priority based on system resource utilization U and task waiting time T, using the formula P = k1*U + k2*T, where k1 and k2 are adjustment coefficients. A higher P indicates a higher priority for the task. The parameter U represents resource utilization, typically ranging from 0 to 1, with the optimal value depending on the specific application scenario. The parameter T represents the task waiting time, usually measured in milliseconds or seconds, and has no fixed optimal value. The coefficients k1 and k2 are used to balance the influence of both and can be adjusted according to actual needs. The main purpose of this formula is to achieve reasonable task scheduling by comprehensively considering resource usage requirements and task urgency, thereby improving system response speed and execution efficiency.

[0115] In one embodiment, one or more computing service nodes are allocated to the real-time streaming task processing queue based on the load value and the number of load tasks of the computing service nodes, specifically including:

[0116] Obtain the prediction execution time of each first prediction network service task in the real-time streaming task processing queue;

[0117] Obtain the predicted load value R and the number of predicted load tasks N for each computing service node at each predicted execution time;

[0118] The computational service node with the smallest value of W = α*R + β*N is selected and assigned to the corresponding first prediction network service task, where α and β are the weight coefficients of R and N, respectively.

[0119] In this embodiment, the dynamic load balancing strategy pre-selects computing service nodes for predicted network service tasks, especially selecting the computing service node with the lowest load when the predicted task is executed. That is, it is necessary to predict the load of the computing service node at a future time point, and the load takes into account both the load value and the number of load tasks.

[0120] For example, such as Figure 4 The dynamic load balancing strategy shown may include the following steps:

[0121] S201, Monitor the real-time load status of the system to identify high-concurrency access requests. This step involves determining the current system load, especially under high-concurrency access conditions, by monitoring various performance metrics (such as CPU utilization, memory utilization, network traffic, etc.) in real time. This data can be collected and processed through a big data platform. For example, in one embodiment, the system collects various performance metrics every 10 seconds and stores this data in a time-series database for rapid analysis and response.

[0122] S202. Allocate different types of processing nodes based on request type and priority. This step assigns requests to the most suitable processing nodes based on the user's request type (e.g., query, write, file upload, etc.) and request priority (high, medium, low). For example, high-priority requests may be assigned to more powerful nodes to ensure a fast response. During this process, the system utilizes AI technology to classify and prioritize requests. Specifically, machine learning models can predict user behavior and needs, thereby intelligently allocating resources.

[0123] S203. Dynamically adjust the task allocation of each processing node using a load balancing algorithm (e.g., weighted round-robin, least connections). This step utilizes multiple load balancing algorithms to optimize task allocation, ensuring a balanced workload for each node. For example, in one embodiment, the system may combine a weighted round-robin algorithm and a least connections algorithm. The weighted round-robin algorithm assigns different weight values ​​to each node based on its processing capacity, while the least connections algorithm assigns new requests to the node with the fewest current connections. Using these two algorithms together allows for more flexible handling of different load conditions. Furthermore, the weight values ​​can be dynamically adjusted based on the historical performance of the nodes, thereby further optimizing load allocation.

[0124] S204. Regularly check and optimize load balancing configurations to ensure high system availability and low latency. This step involves periodically evaluating and adjusting the load balancing configuration to adapt to changes in the system and evolving needs. For example, the system can analyze performance data from the past 24 hours at 2 AM daily to identify potential bottlenecks and problem areas. If inefficient processing of a particular type of request is detected, the weights of the relevant nodes can be adjusted or their configurations optimized. Specifically, the system generates optimization suggestions based on performance metric trends and automatically applies these optimizations via automated scripts. This not only helps improve overall system performance but also ensures a continuously improved user experience.

[0125] The dynamic load balancing formula comprehensively considers factors such as CPU utilization, connection count, and request latency, combining the optimal weight value of each factor to calculate the load value of a node. This allows for dynamic adjustment of task allocation based on the actual load of the nodes, ensuring the system can still operate efficiently under high load conditions. The optimal weight values ​​can be obtained through training with historical data; for example, in a specific implementation, an optimal combination of weight values ​​can be fitted using historical data.

[0126] like Figure 5 The image shows another dynamic load balancing strategy. This strategy aims to maintain high service quality and user satisfaction under different load conditions by adaptively adjusting resource allocation and task processing mechanisms in network services. Specifically:

[0127] S301, the dynamic load balancing strategy automatically adjusts the number of processing nodes and resource allocation based on the current system load. This means that when the system load is high, the number of active processing nodes can be increased to distribute the workload; while when the load is light, unnecessary active nodes are reduced to save energy and improve resource utilization efficiency. For example, an online video conferencing platform can start more servers during peak meeting times to distribute traffic, and shut down redundant servers during off-peak times to reduce operating costs.

[0128] S302, this strategy also adjusts the weight parameter W in the load balancing algorithm based on real-time load conditions. The specific calculation method for the weight parameter W is W = α*R + β*N. Here, R represents a metric of the remaining processing capacity of the current node, ranging from 0 to the maximum available processing capacity; N measures the number of tasks currently being handled by the node, generally a positive integer. α and β are used as weight coefficients for R and N, respectively, to adjust the degree of influence of these two metrics on the overall weight W. The optimal value in an ideal situation depends on the specific circumstances. Through this mechanism, it can be ensured that nodes with strong processing capacity but relatively few tasks are prioritized for scheduling. In a specific example, suppose a computing node X currently has 70% remaining CPU utilization and only 3 unfinished tasks running, while another node Y has 90% CPU reserve but more than 20 tasks waiting to be executed. According to the above formula and after setting an appropriate weight ratio, it may be found that X's actual overall evaluation score is higher, so new tasks will be prioritized for X rather than Y.

[0129] S303 uses historical load data to predict short-term system load trends, allowing for proactive resource allocation adjustments. For example, cloud computing platforms can use machine learning models to study the patterns of user request increases and decreases during past holiday periods. Based on this, they can predict that servers will face greater pressure during upcoming shopping festivals and then increase physical resources or optimize existing virtual cluster configurations in advance to cope with the expected peak demand.

[0130] S304 performs a priority assessment procedure for each new service request that is about to be received. Based on different criteria such as business type, importance level, or potential user impact, it determines whether the request should be processed immediately or temporarily cached and responded to during a quieter period. Taking an online customer service system as an example, if it is an important feedback message marked with an urgent label, it will usually be given high priority and immediately assigned to a human agent for handling; conversely, its urgency can be appropriately reduced until the system load decreases.

[0131] In one implementation, the computing service node prepares to invoke network service resources for network service requests based on several predicted network service tasks, specifically including:

[0132] Obtain the pre-execution time of each first prediction network service task before each prediction execution time, in order to call network service resources;

[0133] The computing service node obtains the stored data that each first prediction network service task needs to access;

[0134] The computing service nodes will call the required stored data to the hot data layer during their respective pre-execution times.

[0135] In this embodiment, preparing to invoke network service resources includes pre-loading the service data required by the predicted network service task to the hot data layer so that the data can be quickly obtained when the task is finally triggered.

[0136] For example, such as Figure 10 As shown, network service providers typically employ tiered storage strategies to efficiently manage large-scale data. The process of using a tiered storage strategy for efficient management of large-scale data includes the following steps: S801, dividing data into three layers for storage: hot data, warm data, and cold data; S802, automatically adjusting the data storage layer based on the data access frequency; S803, periodically auditing storage resources, identifying and releasing unused storage space. This embodiment's method, based on this technology, enables pre-calling of data needed for predicted network tasks, i.e., pre-calling this data to the hot data layer. The balance point between data access frequency H and storage cost C is calculated using the formula Q = lambda * H + mu * C, where lambda and mu are weight parameters, and Q is the optimization objective function.

[0137] The S801 categorizes data into three tiers based on access frequency: hot data, warm data, and cold data. Hot data refers to frequently accessed data, stored on high-performance storage devices to ensure fast access speeds. Warm data is accessed occasionally and stored on storage media that offer slightly lower cost-effectiveness but still support high access efficiency. Cold data consists of rarely accessed archived data, primarily stored on lower-cost but slower-reading storage media. This approach effectively reduces the total cost of ownership of the storage system while ensuring high performance for critical business data.

[0138] S802 automatically adjusts the storage tier of data based on its access frequency. This process monitors data access within the system, and when a change in the access frequency of a certain piece of data is detected, it is automatically moved to the appropriate storage tier. For example, if a user frequently views a log file from a month ago within a certain period, this file, which was originally considered cold data, will be automatically migrated to a warm or even hot data tier due to the sudden increase in access frequency, in order to respond to the user's access requests more quickly. This dynamic data migration not only improves data access efficiency but also makes efficient use of storage resources at different tiers.

[0139] To further improve storage resource utilization and reduce unnecessary expenses, S803 periodically audits storage resources. Audits include, but are not limited to, identifying data blocks that have not been accessed for a long time, duplicate data copies, and expired data. These operations identify and release unused storage space. Specifically, this can be performed as a background task, configured to perform a full scan monthly to identify data that has not been accessed for three consecutive months, and then delete or migrate it to lower-level storage media after appropriate backups.

[0140] Finally, to more scientifically determine how data should move between different storage tiers, the formula Q = lambda * H + mu * C is introduced. Here, H represents the data access frequency, an indicator reflecting how often data is accessed; C represents storage cost, measuring the cost resources consumed in storing data at a specific tier. The weighting parameters lambda and mu control the importance of data access frequency and storage cost in the overall optimization objective, respectively. Typically, lambda ranges from 0.6 to 0.8, indicating a greater emphasis on data access efficiency; while mu is between 0.2 and 0.4, emphasizing that while cost should be considered, service quality should not be completely sacrificed. The formula aims to find the optimal balance between maximizing data value and minimizing operating costs. Through analysis of a large amount of historical data, lambda = 0.75 and mu = 0.25 are found to be a relatively ideal parameter setting scheme, ensuring a good query experience for most users while significantly saving storage costs.

[0141] In addition, the method in this embodiment also includes, Figure 6 The data format conversion and standardization module shown automatically parses and unifies the format of multi-source heterogeneous data, specifically including:

[0142] S401, this process receives heterogeneous data from different sources, performs preliminary parsing, and extracts key fields. For example, in a proactive network service awareness system, this heterogeneous data may include log data from user terminals, network performance data from operators, and API (Application Programming Interface) responses from third-party applications. During the preliminary parsing phase, key fields such as user ID, access time, device type, and request response time are identified and extracted.

[0143] S402, based on predefined data standards, constructs a mapping table and converts extracted fields into a standard format. For example, different data sources may use different timestamp formats (such as UNIX timestamps and ISO 8601 format). After constructing the mapping table, all timestamp fields will be uniformly converted to ISO 8601 format. Furthermore, if the user ID name is inconsistent across different sources, it will also be unified into the `user_id` field through the mapping table.

[0144] S403 uses regular expressions or parsing scripts to process data with special formats. For example, some user feedback may contain unstructured text information. To extract useful information, regular expressions can be written to match and extract keywords, such as network latency, slow loading, etc., and then converted into standardized tags or values.

[0145] S404 checks whether the converted data conforms to a predetermined standard. If it does not, an exception handling mechanism is triggered. The formula is C' = f(D', S'), where D' represents the original data field, S' represents the standard field format, and C' is the validation result. A true C' indicates that the data conforms to the standard. In this formula, parameter D' can be any type of original data field, ranging from strings to numbers; S' is a predefined standard field format, which can be a fixed format or a specific data type. The validation function f compares D' and S' according to specific standardization rules to ensure that the converted data conforms to the expected standard. For example, a timestamp field must be converted to ISO 8601 format, and only if the format is correct can C' be considered true. If the converted data format is incorrect, C' is false, triggering an exception handling mechanism such as logging and alarm notifications. This setting ensures data consistency and reliability, facilitating subsequent analysis and processing. The optimal value is that C' is always true, indicating that all data has been successfully standardized and no anomalies have occurred.

[0146] Overall, such as Figure 4-10 The method shown in the example can solve the following problem:

[0147] 1) Under high concurrency access, sudden surges in data flow can lead to system response latency issues: This paper proposes a dynamic load balancing strategy to optimize big data processing. The main characteristics of big data are large volume, diverse data types, low data value density, and rapid data generation and processing. Key technologies for big data include data acquisition and processing. While traditional dynamic load balancing strategies can adjust dynamically based on real-time server load, their implementation is complex, requiring numerous monitoring and adjustment mechanisms, placing significant pressure on the server. Therefore, under high concurrency access, sudden surges in data flow can cause system response latency, impacting user experience. The dynamic load balancing strategy proposed in this example effectively improves this problem by monitoring the real-time system load status to identify high-concurrency access requests; allocating different types of processing nodes based on request type and priority; dynamically adjusting the task allocation of each processing node using load balancing algorithms (such as weighted round-robin, least connections, etc.); and periodically checking and optimizing the load balancing configuration to ensure high system availability and low latency. This example's dynamic load balancing strategy automatically adjusts the number of processing nodes and resource allocation based on the system's current load. It also adjusts the weight parameter W in the load balancing algorithm based on real-time load conditions; uses historical load data to predict short-term system load trends, thus preparing appropriate resource configuration adjustments in advance; and for each new service request, a priority evaluation process is executed, determining whether the request should be processed immediately or temporarily cached and responded to during idle periods based on different criteria such as business type, importance level, or potential user impact. This strategy dynamically adjusts task allocation among different servers based on the system's real-time load, ensuring that no single node is overloaded by excessive requests, effectively reducing system response latency caused by sudden data surges. When increased pressure is detected on a server, the system can automatically divert some requests to other servers with lower pressure, maintaining the smooth operation of the entire system.

[0148] 2) In the case of multi-source heterogeneous data integration, data parsing errors arise due to inconsistent data formats: The application of a data format conversion and standardization module optimizes big data analysis and processing. While traditional multi-source heterogeneous data integration techniques can extract more details from raw data not available at other levels, they require heavy computational burdens, poor real-time performance, and require strong fault tolerance to handle the instability and uncertainty of the data itself, especially in cases of inconsistent data formats leading to data parsing errors. Differences in data source, semantics, quality, security, privacy, and access permissions can all cause data parsing errors. The data format conversion and standardization module proposed in this example effectively solves the drawbacks of traditional data integration methods. Its process receives heterogeneous data from different sources, performs preliminary parsing and extracts key fields; based on predefined data standards, it constructs a mapping table and converts the extracted fields into a standard format; it uses regular expressions or parsing scripts to process data with special formats; and it determines whether the converted data conforms to the established standards, triggering an exception handling mechanism if it does not. Using the data format conversion and standardization module, it automatically parses data from multiple different sources and formats and converts them into a unified standard format. This module possesses highly intelligent parsing capabilities, adaptively identifying and transforming data structures from different data sources, thus avoiding parsing errors caused by inconsistent formats during data processing. This not only improves the accuracy and efficiency of data processing but also lays a solid foundation for subsequent data analysis.

[0149] 3) In real-time analysis, insufficient data processing speed leads to delays in analysis results: Real-time stream processing technology accelerates the data processing flow. With the help of real-time stream processing technology, the system can process and analyze large amounts of streaming data at a significantly faster speed, greatly shortening the time cycle from data acquisition to the generation of analysis results. Here, "real-time" refers to the simultaneous acquisition of one data packet and processing of another. The traditional data processing flow involves acquiring data and then processing it. This is a serial process, meaning the next step doesn't begin until the previous one is finished. While the data acquisition method is relatively simple, its disadvantages are obvious: serial processing is inefficient, and if an error occurs in one step, it cannot prevent the next step from starting, easily causing system interruptions and freezes. The new data processing workflow is a real-time streaming data processing workflow. It monitors the data stream in real time, detecting any anomalies or bottlenecks during processing; it collects system resource utilization in real time for dynamic resource allocation adjustment; and it optimizes the processing logic based on real-time data feedback to improve overall processing efficiency. This can be understood as the first data packet being processed synchronously in another function while the second packet is being collected. Data acquisition and processing are conducted in parallel, which, while placing higher demands on the function algorithms, significantly shortens the data acquisition and processing time. This is suitable for time-critical fields such as finance. Furthermore, with increasing budgets, multiple algorithm functions can be used to simultaneously collect and process multiple data packets or various data types. For example, 10 algorithm functions can be used to simultaneously collect and analyze 5 data packets or 5 different types of data. This not only greatly improves efficiency but also effectively reduces the probability of system errors or crashes.

[0150] 4) In situations where user behavior changes dynamically, the low training frequency of the model leads to a decrease in prediction accuracy: The system uses a continuous online learning mechanism to update the user behavior prediction model. Traditional mechanisms store user behavior data in a database and then use backend administrators to filter and predict user behavior. This process can also be called learning, but the learning object is the backend administrators. If there are enough administrators or a small number of users, this learning mechanism can meet the needs. However, as the number of users increases, or if low cost is required, this learning mechanism becomes outdated, and errors from manual filtering lead to low prediction accuracy. The new online learning mechanism collects user interaction data in real time, including user clicks, browsing, and search history. This step involves acquiring the latest user behavior information to ensure the model can capture users' latest preferences and behavioral patterns. A database is established to store user behavior data, and this data is then analyzed. The data storage and analysis process is based on big data and deep learning. Both data storage and analysis utilize embedded algorithms to learn and further predict user behavior. The learning object here is the algorithm within the computer, and the learning process can be understood as deep learning or AI. This greatly improves data analysis and processing efficiency and reduces costs. Simultaneously, data filtering and processing through deep learning can significantly improve the accuracy of user behavior prediction. This data is then used to incrementally learn the existing user behavior model online. This is achieved by continuously updating the model's parameters, allowing the model to adapt to the latest user behavior patterns. The model's predictive performance is periodically evaluated, and model parameters are adjusted based on the evaluation results. Evaluation typically involves comparing the model's predictions with actual user behavior to measure the model's accuracy and effectiveness. If the evaluation results show a decline in model performance or room for improvement, the model parameters will be adjusted accordingly. User behavior habits frequently change, and a single static model often struggles to adapt to these changes. Therefore, by adopting an online learning mechanism, the system can regularly train and update the model based on the latest user behavior data, thereby continuously improving the model's prediction accuracy.

[0151] 5) In large-scale data storage scenarios, high storage costs lead to resource optimization issues: applying a tiered storage strategy to optimize large-scale data management. To efficiently manage large-scale data and optimize storage costs, a tiered storage strategy was adopted. Traditional data storage methods include: one is to create a database and store data directly in it, which is simple and fast, but data retrieval and access become slow as the amount of stored data increases, and data retrieval requires filtering large amounts of data, resulting in low efficiency; the other is to create multiple databases, each with different structures or data types, to differentiate and filter the stored data, thereby improving data retrieval efficiency, but its disadvantages are obvious: the creation of multiple databases requires a huge amount of storage space, which leads to a significant increase in costs; in addition, the different reading and storage frequencies and function call frequencies of different data types can easily lead to wasted storage space in some databases, resulting in serious waste of database resources. Compared to traditional data storage methods, the new data storage strategy establishes a database and hierarchically divides it, similar to the second traditional database establishment method. However, this only uses a single database, and function calls are only made to that database, resulting in faster algorithm speeds and significant savings in storage space. This also increases data storage and retrieval efficiency and reduces costs. Furthermore, the algorithm embeds functions based on data importance and access frequency, efficiently classifying data directly during data storage operations. The classification results determine the storage capacity between different layers and media. For example, if layer 1 contains dates with a daily access frequency of 10, and layer 2 contains money with a daily access frequency of 50, the algorithm will set the capacity of layer 1 to be 50 for layer 2. The access frequency between different data types and structures determines the storage capacity of their respective layers, greatly optimizing storage space and resource allocation. This optimized storage space, in turn, improves data read and write speeds, further enhancing the overall system efficiency.

[0152] The following is combined Figure 11-15 Expand explanation Figure 3 Another specific example shown further illustrates the method for proactively perceiving and improving satisfaction with network services based on big data and AI.

[0153] like Figure 11 As shown, the specific steps for intelligently distributing high-concurrency access requests using a dynamic load balancing strategy include:

[0154] S901, Collect real-time load information from each server node. This step involves obtaining the status information of the tasks currently being processed by each server node, including but not limited to CPU utilization, memory usage, network bandwidth usage, and the number of tasks currently in progress. For example, in one embodiment, the system can ensure the real-time nature and accuracy of load data by polling the monitoring metrics of all server nodes every minute.

[0155] S902 calculates the load score for each server node based on the collected load information. The load score is a comprehensive indicator used to measure the current processing capacity of a node, typically comprising a weighted average of multiple sub-indicators. The specific scoring formula can be expressed as: \[\text{Load Score}=w_1\cdot\text{CPU Utilization}+w_2\cdot\text{Memory Usage}+w_3\cdot\text{Number of Tasks}-w_4\cdot\text{Percentage of Remaining Disk Space}\], where \(w_1,w_2,w_3,w_4\) are pre-defined weights, ranging from 0 to 1, and their sum is 1. These weights can be adjusted according to different application scenarios, and the optimal values ​​are usually determined through extensive testing before deployment. For example, in resource-intensive applications, more attention may be paid to CPU utilization and memory usage, so \(w_1\) and \(w_2\) can be set higher, while the other two items can be set lower. The formula is designed this way to comprehensively consider various aspects of server performance, making the scoring results more scientific and reasonable, and accurately reflecting the actual availability of nodes.

[0156] S903 prioritizes allocating new requests to the server node with the highest load rating to achieve load balancing. Specifically, the system maintains a priority queue or ordered set, storing all node information in descending order of load rating. When a new access request is received, the scheduling module directly distributes the request to the server node at the top of the queue (highest rating). This step helps to effectively utilize the most powerful servers in the cluster, ensuring optimal overall system performance. For example, suppose an e-commerce platform is hosting a large-scale promotional event, and user traffic suddenly surges. Through the above steps, some traffic can be quickly directed to the server node with the best performance, thereby ensuring the platform's overall stability and rapid response.

[0157] S904 periodically reassesses the load scores of each node and dynamically adjusts the allocation strategy. To cope with potential surges in data flow, the load scores of nodes need to be updated regularly, with the interval ranging from several minutes to several hours, depending on business needs. Once a new node is detected joining or some nodes fail, the system will immediately trigger a recalculation of the scores and adjust the request allocation strategy to ensure optimal load balancing across the entire system. Specifically, in the e-commerce platform example mentioned above, even after a major promotional event ends and traffic returns to normal, or even if some servers need to be temporarily taken offline for maintenance, this mechanism can promptly detect changes, quickly restore the system's load balancing state, and continue to provide high-quality service.

[0158] like Figure 12 As shown, the specific steps for collecting real-time load information for each server node include:

[0159] S1001, periodically obtain CPU utilization, memory utilization, and network bandwidth utilization from each server node. The purpose of this step is to ensure that the monitoring system can understand the operating status of each server node in real time, thereby promptly identifying potential problems. The specific cycle can be set according to the actual needs of the system, generally ranging from a few seconds to a few minutes. For example, in one embodiment, the system collects these key indicator data from each server node every 10 seconds via the SNMP (Simple Network Management Protocol) protocol.

[0160] S1002, the acquired real-time data is stored in the central monitoring system. The purpose of this step is to centrally manage the data from all server nodes for subsequent analysis and processing. The central monitoring system can be a high-performance database or a distributed file system capable of quickly storing and retrieving large amounts of data. Specifically, a time-series database such as InfluxDB can be used to efficiently store this data, facilitating subsequent querying and analysis.

[0161] S1003, predict server load in the near future based on historical and real-time data. The purpose of this step is to predict server load over future periods using machine learning models, allowing for proactive optimization. For example, time series forecasting algorithms such as ARIMA (Autoregressive Integrated Moving Average Model) or LSTM (Long Short-Term Memory) can be used, combining historical and current real-time data to predict load over the next 15 minutes, thus achieving more accurate resource scheduling.

[0162] S1004, Calculate the load score. The load score formula is \(S=\alpha C+\beta M+\gamma B\), where \(S\) represents the load score, \(C\) represents CPU utilization, \(M\) represents memory utilization, and \(B\) represents network bandwidth utilization.

[0163] (alpha, beta, gamma) are the corresponding weighting coefficients. The values ​​of these coefficients typically range from 0 to 1, with a total sum of 1. The optimal values ​​can be adjusted based on specific business scenarios and actual data. The purpose of this formula is to comprehensively consider the impact of multiple indicators on server load, ensuring that the score fully reflects the current server stress level. For example, in a typical internet application, (alpha = 0.4), (beta = 0.3), and (gamma = 0.3) can be set to emphasize the importance of CPU utilization in the score, as CPU usage is usually one of the key factors affecting performance. The rationale for this formula is that by reasonably allocating weights, the most critical performance bottlenecks can be quickly identified when different resource bottlenecks occur, allowing for effective optimization measures to be taken.

[0164] like Figure 13 As shown, the specific steps for obtaining real-time data through the central monitoring system include:

[0165] S1101, the real-time data collected from each node is formatted into a standard JSON format. This step aims to ensure data consistency and parsability, enabling the central monitoring system to efficiently process and analyze the data. For example, in a big data and AI-based approach to proactively perceive and improve satisfaction in network services, nodes may originate from different devices and sensors, generating data in various formats. By formatting this data into a standard JSON format, the central monitoring system can easily read and process it.

[0166] S1102, the formatted data is transmitted to the central monitoring system. This step involves network communication technologies, such as HTTP (Hypertext Transfer Protocol), MQTT (Message Queuing Telemetry Transport), or TCP (Transmission Control Protocol), to ensure reliable data transmission from each node to the central monitoring system. In one embodiment, the data acquisition module on each node can be configured to send data to the data receiving interface of the central monitoring system periodically or in real time. This transmission mechanism reduces the risk of network congestion and data loss, and improves data integrity and timeliness.

[0167] S1103. After receiving the data, the central monitoring system archives the data based on a preset timestamp. This step ensures the timeliness and orderliness of the data, facilitating subsequent data retrieval and analysis. Specifically, the central monitoring system assigns a unique identifier to each data entry and records its timestamp, then sorts and stores the data according to the timestamp. This method allows users to quickly retrieve data within a specific time range, supporting various historical data analyses and trend predictions.

[0168] S1104, the data storage strategy is based on a time window \(t\). Data within \(t\) retains high resolution, while data exceeding \(t\) undergoes downsampling to optimize storage space. The time window formula is \(t = 3600 times d\), where \(t\) represents the time window length in seconds, and \(d\) represents the number of days the data is retained. This formula aims to effectively utilize storage resources while ensuring high accuracy for recent data. Typically, the parameter \(d\) can range from 1 to 90 days, with the optimal value depending on the specific application's data retention requirements. For example, for a real-time monitoring system, \(d = 7\) can be set, meaning data from the most recent 7 days is stored at high resolution, while data exceeding 7 days is downsampling. Downsampling can be achieved by aggregating multiple data points into an average or maximum value, thus preserving important historical information while reducing storage burden. This approach not only improves system storage efficiency but also makes data management and querying more convenient.

[0169] like Figure 14 As shown, the specific steps for transmitting formatted data to the central monitoring system include:

[0170] S1201, use the HTTPS secure transmission protocol for data transmission to ensure data security. The key to this step is using HTTPS (Hypertext Transfer Protocol Secure), an encrypted transmission protocol that prevents data from being stolen or tampered with by third parties during transmission, thus ensuring data security and integrity. For example, in one embodiment, user behavior data and system log data collected from network services are sent to a central monitoring system via HTTPS to prevent the leakage of sensitive information.

[0171] S1202, Data is sent in batches, with each batch not exceeding 1MB in size. The main purpose of this step is to optimize transmission efficiency and reduce reliance on network bandwidth. By splitting large amounts of data into smaller batches for transmission, transmission failures or network congestion caused by excessively large single data transmissions can be avoided. For example, specifically in proactive network service awareness, user interaction data and system operation status data collected hourly are packaged into multiple batches of less than 1MB each, and then transmitted one by one.

[0172] S1203, data compression is performed before data transmission to reduce transmission time and bandwidth usage. Data compression can significantly reduce the time and bandwidth requirements for data transmission, especially when processing large-scale data. Compressed data occupies less storage space and network resources, helping to improve transmission efficiency. For example, in one embodiment, by using the GZIP algorithm to compress the collected data, the data size was reduced from the original 10MB to approximately 2MB, greatly reducing the time and bandwidth required for transmission.

[0173] S1204, After transmission is complete, the central monitoring system confirms the reception status and returns an acknowledgment message. The acknowledgment message contains a formula to verify transmission integrity: \(H = \text{SHA256}(D)\), where \(H\) represents the SHA256 hash value of the data, and \(D\) represents the transmitted data. The purpose of this formula is to verify whether the data has been modified or corrupted during transmission, ensuring data integrity and accuracy. In this formula, \(H\) is a 256-bit hash value determined by the SHA256 algorithm, while \(D\) is the actual transmitted data. In practical applications, the central monitoring system calculates the SHA256 hash value of the received data and compares it with the hash value provided in the acknowledgment message. If they match, the data transmission is successful and error-free; otherwise, a problem occurred during data transmission, requiring retransmission. This mechanism not only improves the reliability of data transmission but also provides an accurate foundation for subsequent data processing and analysis. For example, in one specific embodiment, the network service proactive sensing system receives an acknowledgment message containing a SHA256 hash value after each batch of data transmission is completed, and performs verification to ensure that the data transmitted each time can be accurately used for further analysis and decision-making.

[0174] like Figure 15 As shown, the specific steps for data archiving based on a preset timestamp include:

[0175] S1301 defines multiple timestamp levels in the central monitoring system. This step aims to categorize data based on its freshness and historical depth, ensuring that data of varying importance is effectively managed and utilized within appropriate timeframes. These levels can be set according to actual needs; for example, data from the most recent day could be at the highest level, with levels decreasing progressively each month.

[0176] S1302, Based on the importance and query frequency of the data, store the data in tiers. This step matches the defined tiers with the specific data to ensure that the data is stored in the most appropriate tier. Important data is usually queried frequently, so it should be placed in a higher tier for faster access; conversely, less important data or data that is rarely queried is stored in a lower tier. This process not only improves data access efficiency but also optimizes the utilization of storage resources.

[0177] S1303, Periodically merge and compress low-level data to free up storage space. This operation is typically performed periodically in the background, reducing the physical footprint of data by merging similar data fragments and thus freeing up more storage space. This helps extend the lifespan of the storage system, reduce costs, and maintain stable system performance. For example, in one embodiment, the system might automatically merge and efficiently compress data from the past year monthly, ensuring that this historical data can continue to be stored without excessive space consumption.

[0178] S1304, the data archiving formula is \(A = L \times Q\), where \(A\) represents the data archiving level, \(L\) represents the data's hierarchical level (ranging from 1 to 10, with higher values ​​indicating higher hierarchical levels), and \(Q\) represents the data's query frequency (ranging from 1 to 100, with higher values ​​indicating higher query frequency). The optimal value depends on the specific application scenario, but generally, higher \(L\) and \(Q\) will result in a higher archiving level \(A\), because higher data importance and query frequency necessitate higher archiving priority. This formula reflects the direct relationship between data archiving level and data importance and query frequency, ensuring that high-value data is processed first. For example, in a specific embodiment, if a user activity data has a hierarchical level \(L\) of 7 and a query frequency \(Q\) of 50, its archiving level \(A\) is calculated to be 350, meaning that this data requires a higher archiving priority to ensure rapid access and processing.

[0179] In summary, through these steps and formulas, the central monitoring system can efficiently manage massive amounts of data, optimize storage and access performance, thereby improving the overall performance of proactive perception and satisfaction of network services based on big data and AI.

[0180] In addition, the specific steps for storing data in layers include:

[0181] First, the data is divided into two main levels: real-time data and historical data. The primary purpose of this step is to optimize storage costs and access efficiency. Real-time data refers to recently generated data that requires frequent access and processing; historical data is data generated earlier and accessed less frequently. In one embodiment, real-time data could be a user's online behavior data from the past few days, while historical data could be a user's behavior records from the past few months or years.

[0182] Next, real-time data is stored in high-performance storage media, such as solid-state drives (SSDs). This layer is characterized by fast access speeds and low latency, making it suitable for high-frequency read and write operations. At this layer, the system uses AI algorithms to analyze the data in real time to achieve proactive awareness of network services. For example, the system can use machine learning models to detect user behavior patterns within network services, promptly identify potential problems, and provide improvement measures.

[0183] Subsequently, historical data is stored on larger-capacity but slower-access storage media, such as hard disk drives (HDDs) or tape libraries. This tier is designed to reduce costs and ensure long-term data retention. During this process, historical data is archived periodically to avoid consuming real-time data storage resources. Specifically, the system can automatically migrate eligible historical data to this storage tier based on preset time windows (e.g., monthly).

[0184] Furthermore, to further optimize storage and query performance, a metadata management mechanism has been introduced. Metadata includes information such as data timestamps, types, and sizes, which is used to accelerate data retrieval and management. For example, when the system needs to extract user behavior records for a specific time period from historical data, metadata can help quickly locate the target data, reducing unnecessary full table scans. Metadata management is achieved through an efficient indexing mechanism, ensuring query efficiency.

[0185] Finally, to ensure data consistency and integrity, a data backup and recovery mechanism was designed. This step includes regularly backing up important data and performing rapid recovery in case of data corruption. For example, the system can perform a full backup daily and incremental backups hourly to ensure that data can be recovered quickly even in the event of an accident. This backup strategy not only improves data security but also reduces recovery time and costs.

[0186] The advantages of the proactive perception and satisfaction improvement method for network services based on big data and AI described in the above example mainly include:

[0187] 1) System Response Optimization under High Concurrency Access: By implementing a dynamic load balancing strategy, high-concurrency access requests are intelligently allocated. This strategy can dynamically adjust the task allocation between different servers based on the real-time load of the system, ensuring that no single node is overloaded by excessive requests, thereby effectively reducing system response latency issues caused by sudden increases in data flow. When an increase in pressure is detected on a certain server, the system can automatically divert some requests to other servers with lower pressure, maintaining the stable operation of the entire system.

[0188] 2) Automatic parsing and standardization of multi-source heterogeneous data: Utilizing a data format conversion and standardization module, the system automatically parses data from multiple different sources and formats, converting it into a unified standard format. This module possesses highly intelligent parsing capabilities, adaptively identifying and converting data structures from different data sources, avoiding parsing errors caused by format inconsistencies during data processing. This not only improves the accuracy and efficiency of data processing but also lays a solid foundation for subsequent data analysis.

[0189] 3) Real-time stream processing technology accelerates data processing workflows: Leveraging real-time stream processing technology, the system can process and analyze large amounts of streaming data at extremely high speeds, significantly shortening the time cycle from data acquisition to analysis result generation. By employing a low-latency data processing architecture and efficient data transmission pipelines, the system can achieve real-time or near-real-time data processing capabilities, thereby meeting the real-time analysis needs of various scenarios and avoiding delays in analysis results caused by insufficient data processing speed.

[0190] 4) Continuous online learning mechanism to update the user behavior prediction model: The system regularly updates the user behavior prediction model through a continuous online learning mechanism. This not only ensures that the model can quickly adapt to dynamic changes in user behavior, but also significantly improves the model's prediction accuracy. By continuously feeding back the latest user behavior data, the system can adaptively adjust the model parameters, thereby making timely predictions when new user behavior patterns emerge, reducing the problem of decreased prediction accuracy caused by low model training frequency.

[0191] 5) Optimizing Large-Scale Data Management with Tiered Storage Strategy: To efficiently manage large-scale data and optimize storage costs, a tiered storage strategy was adopted. This strategy stores data at different levels based on its importance and access frequency. Frequently accessed data is stored on fast-access media, while infrequently accessed data is stored on low-cost, high-capacity storage media. This strategy not only reduces storage costs but also improves data access efficiency and optimizes resource utilization.

[0192] Through the aforementioned series of technological innovations, this disclosure effectively addresses challenges such as high-concurrency access, multi-source heterogeneous data processing, real-time analysis requirements, dynamic changes in user behavior, and large-scale data storage, thereby significantly improving the performance of network services and user experience.

[0193] Example 2:

[0194] like Figure 2 As shown, this disclosure provides a network service apparatus, the apparatus comprising:

[0195] Receiver module 1 is used to receive network service requests from users;

[0196] Prediction module 2, connected to receiving module 1, is used to predict several predictive network service tasks that will be executed due to subsequent user behavior, based on network service requests and user behavior prediction models.

[0197] The allocation module 3, connected to the prediction module 2, is used to allocate computing service nodes to network service requests based on several predicted network service tasks and dynamic load balancing strategies, so that the computing service nodes can prepare to call network service resources for network service requests based on several predicted network service tasks.

[0198] In one embodiment, wherein:

[0199] Network service requests include requests from users to browse products on e-commerce platforms;

[0200] Several predictive network service tasks include browsing products, switching browsing modes, adding items to the cart, and completing a purchase.

[0201] In one embodiment, the prediction module 2 specifically includes:

[0202] The learning unit is used to obtain the first historical user behavior data based on network service requests. The first historical user behavior data includes several historical network service tasks that were executed after the user made the same type of network service request in the past, as well as the historical execution waiting time and historical execution probability of each historical network service task.

[0203] The prediction unit, connected to the learning unit, is used to use the user behavior prediction model to predict, based on the first historical user behavior data, several prediction network service tasks that will be executed by subsequent user behavior, as well as the prediction execution waiting time and prediction execution probability of each prediction network service task.

[0204] In one embodiment, the prediction module 2 further includes:

[0205] The model evaluation unit is used to evaluate whether the user behavior prediction model needs to be updated according to the following formula: F = a*E + b*G, where a and b are balancing factors, E is the model prediction error obtained from the second historical user behavior data, and G represents the model complexity.

[0206] The model update unit, connected to the model evaluation unit, is used to update the user behavior prediction model according to the third historical user behavior data in response to the need for updating the evaluation user behavior prediction model. The third historical user behavior data includes the second historical user behavior data and is longer than the first historical user behavior data.

[0207] In one embodiment, the allocation module 3 specifically includes:

[0208] The real-time streaming unit is used to establish a real-time streaming task processing queue for network service requests. Based on the predicted execution waiting time and the predicted execution probability, several predicted network service tasks are added to the real-time streaming task processing queue.

[0209] The dynamic load balancing unit, connected to the real-time streaming unit, is used to allocate one or more computing service nodes to the real-time streaming task processing queue based on the load value and number of load tasks of the computing service nodes.

[0210] In one embodiment, the real-time streaming unit specifically includes:

[0211] The task selection sub-unit is used to obtain the first prediction network service task whose prediction execution probability is greater than the first threshold among several prediction network service tasks.

[0212] The task requirement subunit, connected to the task selection subunit, is used to obtain the required network service resources when each first prediction network service task is executed.

[0213] The task priority subunit, connected to the task demand subunit, is used to obtain the resource preparation call priority of each first predicted network service task based on the demand network service resource quantity U and the predicted execution waiting time T: P = k1*U + k2*T, where k1 and k2 are adjustment coefficients with opposite positive and negative values.

[0214] The task sorting subunit, connected to the task priority subunit, is used to add each first prediction network service task to the real-time stream task processing queue according to the priority of resource preparation call.

[0215] In one embodiment, the dynamic load balancing unit specifically includes:

[0216] The prediction execution time subunit is used to obtain the prediction execution time of each first prediction network service task in the real-time streaming task processing queue.

[0217] The predicted load subunit, connected to the predicted execution time subunit, is used to obtain the predicted load value R and the number of predicted load tasks N for each computing service node at each predicted execution time.

[0218] The node selection subunit, connected to the prediction load subunit, is used to select the computing service node with the smallest value of W = α * R + β * N and assign it to the corresponding first prediction network service task. α and β are the weight coefficients of R and N, respectively.

[0219] In one embodiment, the allocation module 3 further includes:

[0220] The pre-execution time subunit is used to obtain the pre-execution time of each first prediction network service task before each prediction execution time for calling network service resources;

[0221] The computing service node is connected to the pre-execution time sub-unit and is used to obtain the storage data that each first prediction network service task needs to call, and to call the storage data that needs to be called to the hot data layer at each corresponding pre-execution time.

[0222] Example 3:

[0223] Embodiment 3 of this disclosure provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it implements the network service method as described in Embodiment 1, or the network service apparatus as described in Embodiment 2.

[0224] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program units, or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), DVD or other optical disc storage, cartridges, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer.

[0225] Additionally, this disclosure may provide a computer device including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the network service method as described in Embodiment 1. This computer device may be the network service device as described in Embodiment 2.

[0226] The memory is connected to the processor. The memory can be flash memory, read-only memory or other types of memory. The processor can be a central processing unit or a microcontroller.

[0227] Embodiments 1-3 of this disclosure provide a network service method, a network service device, and a computer-readable storage medium. By predicting user behavior, the subsequent network service tasks triggered by network service requests are obtained. By combining dynamic load balancing and a pre-call resource mode, computing nodes and network service resources are allocated in advance for the predicted network service tasks, which can respond to user needs more promptly and improve the user's network experience.

[0228] It is understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of this disclosure, and this disclosure is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of this disclosure, and these modifications and improvements are also considered to be within the scope of protection of this disclosure.

Claims

1. A network service method, characterized in that, The method includes: Receive network service requests from users; Based on the network service request and user behavior prediction model, predict the network service tasks that will be executed when the user behavior occurs, as well as the predicted execution waiting time and predicted execution probability of each network service task. Based on several predicted network service tasks and dynamic load balancing strategies, computing service nodes are allocated to network service requests. These computing service nodes then prepare to access network service resources for the requests based on the predicted network service tasks. Specifically, this includes: Establish a real-time streaming task processing queue for network service requests. Among several prediction network service tasks, identify the first prediction network service task whose predicted execution probability is greater than a first threshold. Obtain the required network service resources for each first prediction network service task during execution. Based on the required network service resource quantity U and the predicted execution waiting time T, the resource preparation call priority for each first predicted network service task is obtained: P = k1 × U + k2 × T, where k1 and k2 are adjustment coefficients with opposite positive and negative values. According to the priority of resource preparation and allocation, each first prediction network service task is added to the real-time streaming task processing queue. Based on the load value and number of tasks on the computing service node, allocate one or more computing service nodes to the real-time streaming task processing queue. Obtain the pre-execution time of each first prediction network service task before its prediction execution time, in order to invoke network service resources. The computing service nodes obtain the stored data that each first prediction network service task needs to access. The computing service nodes will call the required stored data to the hot data layer during their respective pre-execution times.

2. The method according to claim 1, characterized in that, in: Network service requests include requests from users to browse products on e-commerce platforms; Several predictive network service tasks include browsing products, switching browsing modes, adding items to the cart, and completing a purchase.

3. The method according to claim 1 or 2, characterized in that, Based on the network service request and user behavior prediction model, several predicted network service tasks that will be triggered by subsequent user behavior are predicted, along with the predicted execution wait time and predicted execution probability of each predicted network service task, specifically including: The first historical user behavior data is obtained based on the network service request. The first historical user behavior data includes several historical network service tasks that were executed after the user made the same type of network service request in the past, as well as the historical execution waiting time and historical execution probability of each historical network service task. Using a user behavior prediction model based on first-historical user behavior data, predict several predictive network service tasks that will be executed due to subsequent user behavior, as well as the predicted execution waiting time and predicted execution probability of each predictive network service task.

4. The method according to claim 3, characterized in that, Prior to using the user behavior prediction model, the method further includes: The user behavior prediction model needs to be updated according to the following formula: F = a × E + b × G, where a and b are balancing factors, E is the model prediction error obtained from the second historical user behavior data, and G represents the model complexity. In response to the need to update the user behavior prediction model for evaluation, the user behavior prediction model is updated based on third historical user behavior data, which includes second historical user behavior data and is longer than first historical user behavior data.

5. The method according to claim 1, characterized in that, Based on the load value and number of tasks of the computing service nodes, one or more computing service nodes are allocated to the real-time streaming task processing queue, specifically including: Obtain the prediction execution time of each first prediction network service task in the real-time streaming task processing queue; Obtain the predicted load value R and the number of predicted load tasks N for each computing service node at each predicted execution time; The computational service node with the smallest value of W = α × R + β × N is selected and assigned to the corresponding first prediction network service task, where α and β are the weight coefficients of R and N, respectively.

6. A network service device, characterized in that, The device includes: The receiving module is used to receive network service requests from users. The prediction module, connected to the receiving module, is used to predict, based on network service requests and user behavior prediction models, several predicted network service tasks that will be triggered by subsequent user behavior, as well as the predicted execution waiting time and predicted execution probability of each predicted network service task. The allocation module, connected to the prediction module, is used to allocate computing service nodes to network service requests based on several predicted network service tasks and dynamic load balancing strategies. This allows the computing service nodes to prepare network service resources for the network service requests based on the predicted network service tasks. Specifically, it includes functions for: Establish a real-time streaming task processing queue for network service requests. Among several prediction network service tasks, identify the first prediction network service task whose predicted execution probability is greater than a first threshold. Obtain the required network service resources for each first prediction network service task during execution. Based on the required network service resource quantity U and the predicted execution waiting time T, the resource preparation call priority for each first predicted network service task is obtained: P = k1 × U + k2 × T, where k1 and k2 are adjustment coefficients with opposite positive and negative values. According to the priority of resource preparation and allocation, each first prediction network service task is added to the real-time streaming task processing queue. Based on the load value and number of tasks on the computing service node, allocate one or more computing service nodes to the real-time streaming task processing queue. Obtain the pre-execution time of each first prediction network service task before its prediction execution time, in order to invoke network service resources. The computing service nodes obtain the stored data that each first prediction network service task needs to access. The computing service nodes will call the required stored data to the hot data layer during their respective pre-execution times.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the network service method as described in any one of claims 1-5.

Images