Content scheduling method and device, equipment and storage medium

Abstract

The invention relates to the field of mobile content distribution, and provides a content scheduling method and device, equipment and a storage medium. The method comprises the following steps: acquiring attribute data of to-be-distributed content, state data of at least one target node in a content distribution network and external influence factor data; inputting the attribute data of the to-be-distributed content, the state data of the at least one target node and the external influence factor data into a trained content distribution quantity prediction model to obtain predicted distribution quantity information of the to-be-distributed content on the at least one target node in a future specified time period; and performing content pre-distribution and / or resource adjustment on the at least one target node based on the predicted distribution quantity information. According to the content scheduling method provided by the invention, accurate and fine-grained prediction can be carried out on the content distribution quantity of the CDN nodes, so that the accuracy of content pre-distribution and the utilization efficiency of network resources are improved.

Description

Technical Field

[0001] This invention relates to the field of mobile content distribution technology, and in particular to a content scheduling method, apparatus, device, and storage medium. Background Technology

[0002] Currently, mobile content delivery networks (CDNs) mainly rely on two modes for content scheduling: one is the passive scheduling mode, which sends content from the origin server or upstream node back to the edge node near the user based on the real-time access request initiated by the user; the other is the proactive injection mode based on experience prediction, which pushes content that is expected to be popular to some nodes in advance based on simple historical data analysis or operational experience.

[0003] Current CDN content scheduling methods mainly rely on passive response or experience-based judgment, making it difficult to effectively cope with rapid changes in content popularity and regional differences. Summary of the Invention

[0004] This application provides a content scheduling method, apparatus, device, and storage medium to address the technical problem that existing technologies struggle to effectively cope with rapid changes in content popularity and regional differences.

[0005] In a first aspect, embodiments of this application provide a content scheduling method, including: Acquire attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and data on external influencing factors; The attribute data of the content to be distributed, the status data of the at least one target node, and the data of the external influencing factors are input into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network. Based on the predicted distribution volume information, content pre-distribution and / or resource adjustment are performed on the at least one target node.

[0006] In some embodiments, obtaining attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and current external influencing factor data includes: Obtain at least one of the type, size, and publication time of the content to be distributed as attribute data of the content to be distributed; Obtain at least one of the hierarchical position, geographical location, and bandwidth capacity of the at least one target node as the status data of the at least one target node; At least one of the following external influencing factors is obtained: seasonal information, marketing campaign information, recommendation system information, and social hot topic information.

[0007] In some embodiments, the step of inputting the attribute data of the content to be distributed, the state data of the at least one target node, and the external influencing factor data into a trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node within a specified future time period includes: Output the predicted cumulative distribution volume curve of the content to be distributed on at least one target node over time; And / or, output the predicted peak distribution time and peak distribution amount of the content to be distributed on at least one target node.

[0008] In some embodiments, the pre-distribution of content or adjustment of resources for the at least one target node based on the predicted distribution volume information includes: When the peak distribution volume in the predicted distribution volume information exceeds a preset first threshold, a pre-distribution instruction is generated to actively push the content to be distributed from the upper-level node or the source station to the at least one target node. And / or, when the total predicted distribution amount for the target node in the predicted distribution amount information exceeds a preset second threshold, an adjustment suggestion for bandwidth expansion or load balancing of the at least one target node is generated.

[0009] In some embodiments, the method further includes obtaining the trained content distribution volume prediction model, wherein obtaining the trained content distribution volume prediction model includes: Collect the historical operation dataset of the content distribution network. The historical operation dataset includes historical content attribute data of multiple historical contents, as well as historical distribution volume time-series data, historical node status data, and historical external influencing factor data of each node. Using the historical operation dataset as training samples, the parameters of the initial diffusion model built based on Bass diffusion theory are trained to obtain the trained content distribution volume prediction model and its model parameter set.

[0010] In some embodiments, the first node is any node in the content delivery network, and the initial diffusion model models the distribution process of the first node as follows: The instantaneous distribution change rate of the first node at the target time is equal to the sum of the first and second terms multiplied by the remaining distribution capacity of the first node; The first term is the content push coefficient representing the proactive distribution capability of the first node itself, and the second term is the product of the content request coefficient representing the user request-driven capability of the first node and the current cumulative distribution volume ratio.

[0011] In some embodiments, the initial diffusion model also introduces an inter-node influence coefficient to characterize the probability of content spreading from other nodes to the first node; When modeling the first node, the sum of the first and second terms also includes a third term, which is the sum of the current cumulative distribution ratio of other nodes after being weighted by the influence coefficient between the corresponding nodes.

[0012] In some embodiments, the initial diffusion model further introduces at least one dynamic influence factor function, which is used to dynamically adjust the distribution process of the first node based on the input content attribute data, node status data and / or external influencing factor data.

[0013] In some embodiments, the dynamic impact factor function includes at least one of the following: Importance factor function, used to characterize the decay of content value over time; The relevance factor function is used to characterize the degree of matching between content and user groups; Seasonal factor function, used to characterize the fluctuation of distribution volume over time period; Popularity factor function is used to characterize the growth trend of content popularity.

[0014] In some embodiments, the parameter training is achieved by constructing and minimizing the following objective function: the objective function includes the sum of errors between the predicted distribution amount and the actual historical distribution amount of the initial diffusion model on all training samples, and regularization constraints on the content push coefficient, the content request coefficient and the inter-node influence coefficient.

[0015] Secondly, embodiments of this application provide a content scheduling device, comprising: The data acquisition module is used to acquire attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and data on external influencing factors. The distribution volume prediction module is used to input the attribute data of the content to be distributed, the status data of the at least one target node, and the external influencing factor data into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network. The content scheduling module is used to pre-distribute content and / or adjust resources for the at least one target node based on the predicted distribution volume information.

[0016] Thirdly, embodiments of this application provide an electronic device, including a processor and a memory storing a computer program, wherein the processor executes the program to implement the content scheduling method described in the first aspect.

[0017] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the content scheduling method described in the first aspect.

[0018] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the content scheduling method described in the first aspect.

[0019] The content scheduling method, apparatus, device, and storage medium provided in this application embodiment achieve accurate and fine-grained prediction of the content distribution volume of CDN nodes through a prediction method based on the Bass diffusion model, thereby improving the accuracy of content pre-distribution and the efficiency of network resource utilization. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts illustrating the content scheduling method provided in the embodiments of this application; Figure 2 This is a second flowchart illustrating the content scheduling method provided in the embodiments of this application; Figure 3 This is a flowchart illustrating the CDN node content distribution volume prediction method provided in the embodiments of this application; Figure 4 This is a schematic diagram of the structure of the content scheduling device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the physical structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] Mobile Content Delivery Networks (CDNs) are characterized by their massive content volume, wide user base and coverage, and high request volume. To balance network traffic and improve user experience, CDNs consist of content centers and numerous edge nodes, with these edge nodes extending deep into cities and counties. In the operation of a content network, accurately distributing content of interest to customers to various content centers and edge nodes is crucial for improving the efficiency of content network resource utilization and user experience.

[0024] Currently, mobile content delivery networks (CDNs) mainly rely on two modes for content scheduling: one is the passive scheduling mode, which sends content from the origin server or upstream node back to the edge node near the user based on the real-time access request initiated by the user; the other is the proactive injection mode based on experience prediction, which pushes content that is expected to be popular to some nodes in advance based on simple historical data analysis or operational experience.

[0025] Furthermore, although content networks possess information related to content popularity, they currently do not fully utilize this data. They have not introduced algorithmic models to predict the development trends of content hotspots, thus failing to achieve more proactive and predictive content scheduling. This results in insufficient accuracy and predictability in content scheduling, often leading to excessively long user access delays during peak business periods due to passive content backhaul, which negatively impacts user experience.

[0026] Current CDN content scheduling methods mainly rely on passive response or experience-based judgment, making it difficult to effectively cope with rapid changes in content popularity and regional differences.

[0027] Figure 1 This is one of the flowcharts illustrating the content scheduling method provided in the embodiments of this application. For example... Figure 1 As shown, this content scheduling method may include: Step 101: Obtain the attribute data of the content to be distributed, the status data of at least one target node in the content distribution network, and the data of external influencing factors.

[0028] It should be noted that the available data examples include, but are not limited to: the type, size, and publication time of the content to be distributed; the hierarchical position of the target node (such as content center, provincial node, prefecture-level node, etc.), geographical location, and bandwidth capacity; as well as external influencing factors such as seasonal information, marketing campaign information, recommendation system trigger events, and social hot topics.

[0029] Step 102: Input the attribute data of the content to be distributed, the status data of the at least one target node, and the data of the external influencing factors into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network.

[0030] It should be noted that the trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model. Specifically, this application creatively applies the classic Bass diffusion theory to the CDN field to predict the distribution volume of content throughout its entire lifecycle on CDN nodes. Furthermore, the model incorporates the network structure and node characteristics of the CDN into the diffusion model, developing a hierarchical Bass model. This model can configure independent model parameter sets for different nodes in the content distribution network, thereby modeling and predicting the distribution volume of content on different levels of nodes. The model can also further introduce dynamic variables reflecting content characteristics, network status, and the influence of the external environment, and capture the propagation process of content between nodes through inter-node influence coefficients, thus more accurately describing the dynamics of content distribution in the CDN.

[0031] Step 103: Based on the predicted distribution volume information, perform content pre-distribution and / or resource adjustment on the at least one target node.

[0032] It should be noted that corresponding scheduling or resource management actions can be performed based on the prediction results. For example, if the prediction information indicates that the distribution volume of certain content on a specific node will exceed a preset threshold, proactive content push from the origin server or upstream node to that node can be triggered; if the prediction information indicates that the total distribution load of a node in a future time period will approach its capacity limit, resource expansion or load balancing suggestions for that node can be generated. This step transforms content scheduling from a passive response to proactive optimization based on prediction, thereby improving network resource utilization efficiency and user experience.

[0033] This application combines Bass diffusion theory with CDN network characteristics through the above method, achieving accurate and fine-grained prediction of content distribution trends, thereby significantly improving the initiative, predictability and resource utilization efficiency of CDN content scheduling.

[0034] In some embodiments, obtaining attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and current external influencing factor data includes: Obtain at least one of the type, size, and publication time of the content to be distributed as attribute data of the content to be distributed; Obtain at least one of the hierarchical position, geographical location, and bandwidth capacity of the at least one target node as the status data of the at least one target node; At least one of the following external influencing factors is obtained: seasonal information, marketing campaign information, recommendation system information, and social hot topic information.

[0035] It should be noted that the acquisition of the above data provides the necessary input foundation for the content distribution volume prediction model. Content attribute data reflects the dissemination characteristics of the content itself, node status data reflects the service capabilities and constraints of different nodes, and external influencing factor data is used to construct the dynamic influencing factors in the model.

[0036] By acquiring the aforementioned multi-dimensional data, this method can provide comprehensive and fine-grained input features for prediction based on the Bass diffusion model, thereby enhancing the model's ability to characterize the dynamics of content propagation in complex CDN networks, improving the accuracy of prediction results and their applicability in actual content distribution and scheduling.

[0037] In some embodiments, the step of inputting the attribute data of the content to be distributed, the state data of the at least one target node, and the external influencing factor data into a trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node within a specified future time period includes: Output the predicted cumulative distribution curve of the content to be distributed on at least one target node over time; and / or, output the predicted peak distribution time and peak distribution amount of the content to be distributed on at least one target node.

[0038] It's important to note that, based on the characteristics of the Bass diffusion model, the cumulative distribution volume curve typically exhibits an S-shaped trend. This curve reflects the entire lifecycle of content from initial distribution to growth saturation. Analyzing this curve allows for the calculation of key indicators such as the peak arrival time and peak distribution volume. Specifically, by analyzing the S-shaped curve of content distribution volume, we can predict the peak arrival time and magnitude of content distribution on the CDN, thereby proactively identifying trending content and predicting the duration for which trending content maintains a high distribution volume.

[0039] By outputting the cumulative distribution volume curve and distribution peak information, this method can intuitively display the entire lifecycle distribution trend of content on CDN nodes, enabling operators to identify potential hot content in advance and accurately predict its peak distribution time and magnitude. This predictive capability helps CDN operators prepare resources and schedule arrangements in advance, avoiding service quality degradation caused by sudden hotspots, thereby improving network resource utilization efficiency and user experience.

[0040] For example, the main output of a content distribution prediction model may include: The predicted distribution volume of each level of node is the output of a single piece of content. At each level of node (content center, provincial node, municipal node, district / county node) In different time periods in the future Predicted distribution volume The predicted distribution volume of nodes at each level can guide CDN operators to make accurate resource allocations and avoid node overload or resource waste.

[0041] Total predicted distribution volume per node, which is the total predicted distribution volume for different time periods in the future, aggregated from the predicted distribution volume of individual content on each node at each level. (i.e., node) (cumulative distribution of all content on the network) and the cumulative distribution of content used by the entire content network. The aggregated total predicted distribution volume can help CDN operators assess overall network load, plan capacity, identify potential network bottlenecks, and proactively expand bandwidth or load balance.

[0042] The key parameters obtained from the fitting, such as the content push coefficient α and the content request coefficient β, can help CDN operators gain a deeper understanding of the impact of different factors on content dissemination, thereby providing quantitative basis for content push strategies and user behavior analysis.

[0043] The S-shaped curve of content distribution volume is a curve that plots the trend of individual content distribution volume over time based on a defined model structure and estimated parameters. It visually displays the distribution volume at different stages of the content lifecycle (initial, rapid growth, saturation). The Bass model can quickly provide reasonable predictions based on a small amount of initial data and domain knowledge (model structure), making it particularly suitable for predicting the distribution volume of newly released content.

[0044] Content distribution peak prediction: By analyzing the generated S-shaped curve, the peak distribution time and duration of high distribution volume can be estimated, which can help CDN operators identify potential hot content in advance.

[0045] In some embodiments, the pre-distribution of content or adjustment of resources for the at least one target node based on the predicted distribution volume information includes: When the peak distribution volume in the predicted distribution volume information exceeds a preset first threshold, a pre-distribution instruction is generated to actively push the content to be distributed from the upper-level node or the source station to the at least one target node; and / or, when the total predicted distribution volume for the target node in the predicted distribution volume information exceeds a preset second threshold, an adjustment suggestion for bandwidth expansion or load balancing of the at least one target node is generated.

[0046] It should be noted that the first and second thresholds mentioned above can be dynamically set based on the actual service capacity of the nodes, historical load conditions, and business strategies. This threshold judgment mechanism transforms content scheduling and resource management from passive response to proactive intervention based on prediction. The content distribution prediction model, based on the Bass diffusion model, builds powerful predictive capabilities for CDN content distribution, providing data insights into future content distribution trends and enabling smarter and more efficient operation and management of the content distribution network. Its application scenarios include, but are not limited to: 1. Content distribution strategy optimization: Utilize the predicted distribution volume curve to optimize the pre-distribution of content among nodes, reduce network congestion, and improve user access speed.

[0047] 2. Resource Planning: Based on long-term forecasts, we help CDN operators plan their bandwidth and storage resources, improving resource utilization efficiency and reducing operating costs.

[0048] 3. Hot Topic Content Identification: By predicting peak distribution times, we can identify potentially high-profile content in advance, prepare resources accordingly, and avoid a decline in service quality due to sudden hot topics.

[0049] 4. Cache strategy optimization: Based on the predicted content lifecycle, optimize the content caching time and strategy for different nodes to improve cache hit rate and reduce the pressure on the origin server.

[0050] 5. Cost estimation: Estimate the distribution cost of content throughout its entire lifecycle to assist in pricing decisions and optimize operational efficiency.

[0051] 6. Revenue Forecasting: For paid content, long-term revenue potential can be predicted to support content investment decisions and revenue analysis.

[0052] By combining the prediction results with preset thresholds and business strategies, this method can achieve closed-loop management from prediction to execution, making CDN content scheduling and resource allocation more proactive, accurate and efficient, thereby significantly improving network service quality, resource utilization and operational efficiency.

[0053] Figure 2 This is a second flowchart illustrating the content scheduling method provided in the embodiments of this application. Figure 2 As shown, this content scheduling method may include: Step 201: Obtain the trained content distribution volume prediction model; Step 202: Obtain the attribute data of the content to be distributed, the status data of at least one target node in the content distribution network, and the data of external influencing factors.

[0054] Step 203: Input the attribute data of the content to be distributed, the status data of the at least one target node, and the data of the external influencing factors into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node within a specified time period in the future.

[0055] Step 204: Based on the predicted distribution volume information, perform content pre-distribution and / or resource adjustment on the at least one target node.

[0056] In some embodiments, obtaining the trained content distribution volume prediction model includes: Collect the historical operation dataset of the content distribution network. The historical operation dataset includes historical content attribute data of multiple historical contents, as well as historical distribution volume time-series data, historical node status data, and historical external influencing factor data of each node. Using the historical operation dataset as training samples, the parameters of the initial diffusion model built based on Bass diffusion theory are trained to obtain the trained content distribution volume prediction model and its model parameter set.

[0057] It should be noted that the Bass Diffusion Model (Bass Diffusion Model for short) is a classic model used to predict the adoption rate of new products or technologies, proposed by Frank M. Bass in 1969. This model is based on the theory of innovation diffusion and assumes that adopters of new products or technologies can be divided into two categories: innovators and imitators. The basic form of the Bass model is as follows: .

[0058] in, yes The cumulative number of adopters at any given time; It is market potential (the maximum number of potential adopters); It is the innovation coefficient (reflecting the influence of innovators and representing the probability of independent adoption); It is the imitation coefficient (reflecting the influence of imitators, representing the probability of adoption based on social network influence).

[0059] One important advantage of the Bass model is that it has analytical solutions, specifically: .

[0060] This allows the model to not only perform numerical simulations but also directly calculate the cumulative distribution at any given time, greatly improving computational efficiency and prediction accuracy.

[0061] In CDN content distribution scenarios, the process of content dissemination and consumption can be viewed as a diffusion process, with content within the CDN considered as "innovation," and content distribution within the CDN network seen as the diffusion of innovation. Therefore, based on the Bass model, a content distribution prediction model can be constructed: ; in, It represents the cumulative content distribution volume of a single piece of content at time t; This is the maximum capacity of the node; It is the content push coefficient (proactive distribution); It is the content request coefficient (driven by user demand).

[0062] The Bass model effectively captures the core dynamics of content distribution, requiring only a small number of parameters (mainly the content push coefficient α and the content request coefficient β) to describe the complex distribution process. Its simplicity makes the model easy to understand, implement, and maintain, without sacrificing its predictive power. By integration, an S-shaped curve can be obtained showing the cumulative distribution volume of a single piece of content in the CDN network: ; This S-shaped curve can describe the content from... arrive → The total distribution volume throughout its entire lifecycle. By analyzing the S-shaped curve, we can: Predicted peak content distribution time: Peak arrival time is Peak distribution volume is .

[0063] Calculate the distribution speed: .

[0064] Estimate the time it takes for the cumulative distribution to reach half of the maximum capacity (half-saturation time): .

[0065] The Bass model can describe the entire lifecycle of content distribution, including the initial stage (slow growth mainly driven by innovators), the rapid growth stage (exponential growth dominated by imitation effects), and the saturation stage (gradually slowing growth). It closely matches the typical content distribution pattern in CDN and is particularly suitable for predicting the spread trend of trending content. It is very valuable for CDN's long-term resource planning and strategy formulation.

[0066] By employing a prediction model based on Bass diffusion theory, this method accurately depicts the entire lifecycle diffusion process of content within CDN nodes while maintaining model simplicity and high computational efficiency. This model requires only a small number of parameters with clear business implications (such as the content push coefficient α and content request coefficient β) to achieve rapid and reliable prediction of the distribution trends of newly released and trending content, providing strong data support and theoretical basis for CDN operational decisions such as content pre-distribution, resource planning, and hotspot identification.

[0067] In some embodiments, the first node is any node in the content delivery network, and the initial diffusion model models the distribution process of the first node as follows: The instantaneous distribution change rate of the first node at the target time is equal to the sum of the first and second terms multiplied by the remaining distribution capacity of the first node; The first term is the content push coefficient representing the proactive distribution capability of the first node itself, and the second term is the product of the content request coefficient representing the user request-driven capability of the first node and the current cumulative distribution volume ratio.

[0068] It should be noted that, in order to more accurately describe the content distribution process in CDN, this application incorporates variables reflecting content characteristics and external influences into the basic Bass model. Considering the network structure of CDN, this application develops a hierarchical Bass model to model the diffusion process at different levels (such as content centers and edge nodes). For example, for nodes... The model can be represented as: ; in, , and These are nodes Content push coefficient, content request coefficient, and maximum capacity; It is a content importance variable. It is a content relevance variable. It is a seasonal variable. It is a popularity variable. For node indexing, Define the time variable and its zero point. Represents a node The current saturation of the content. Represents a node The remaining distribution potential of the content. These variables can be dynamically calculated based on content attributes, node status, and external influencing factors, for example, by modeling using exponential decay functions, logistic functions, periodic functions, etc.

[0069] By introducing the aforementioned variables and a hierarchical parameter design (i.e., different nodes can have independent parameters), we can achieve this. , and This method can model the diffusion process of nodes at different levels (such as content centers, provincial nodes, and municipal nodes) based on the network structure and node characteristics of CDN, thereby achieving more granular and accurate predictions of content distribution trends. This modeling approach not only retains the advantages of the Bass model in terms of simplicity and efficiency, but also significantly enhances the model's adaptability to the complex environment of CDN, providing a more reliable theoretical foundation for content pre-distribution and dynamic resource scheduling.

[0070] In some embodiments, the initial diffusion model also introduces an inter-node influence coefficient to characterize the probability of content spreading from other nodes to the first node; When modeling the first node, the sum of the first and second terms also includes a third term, which is the sum of the current cumulative distribution ratio of other nodes after being weighted by the influence coefficient between the corresponding nodes.

[0071] It should be noted that, in order to capture content propagation between nodes, an inter-node influence coefficient is introduced. , indicating that the content starts from the node propagation to nodes The probability of: ; This extended model considers content propagation between nodes, making predictions more accurate. To further improve prediction accuracy, this application defines a global objective function to optimize the model parameters: ; in, It is a node The actual distribution amount at time t This is the regularization parameter. This optimization problem can be solved using methods such as gradient descent. With this optimized, complex model, the content distribution process in a CDN can be predicted more accurately, providing strong support for content scheduling and resource allocation.

[0072] By introducing an inter-node influence coefficient and optimizing the objective function, this method can significantly improve the prediction accuracy of the content propagation process in the CDN network, providing more reliable data support for content scheduling and resource allocation.

[0073] In some embodiments, the initial diffusion model further introduces at least one dynamic influence factor function, which is used to dynamically adjust the distribution process of the first node based on the input content attribute data, node status data and / or external influencing factor data.

[0074] Furthermore, the dynamic influence factor function includes at least one of the following: Importance factor function, used to characterize the decay of content value over time; The relevance factor function is used to characterize the degree of matching between content and user groups; Seasonal factor function, used to characterize the fluctuation of distribution volume over time period; Popularity factor function is used to characterize the growth trend of content popularity.

[0075] It should be noted that, in order to more accurately describe the content distribution process in CDN, this application adds variables reflecting content characteristics and external influences to the basic Bass model, and the optimized content distribution prediction model is expressed as: ; in, yes The importance of the content at any given moment; yes Content-related variables at any given time; yes Seasonal variables at any given time; yes Popularity variable at any given moment.

[0076] These factors can be represented in different functional forms, for example: Content importance factor: ,in It is the initial importance. It is the attenuation rate.

[0077] Content relevance factors: This is a Logistic function. Control the steepness of the curve. It's an inflection point.

[0078] Seasonal factors: ,in It is the amplitude. It is a cycle.

[0079] Popularity factor: ,in It is the most popular trend. It is the growth rate.

[0080] This application introduces variables reflecting content characteristics and external influences such as seasonality and marketing activities into the Bass model, optimizing the Bass diffusion model to make it more suitable for CDN scenarios. This allows it to more accurately depict the complex dynamics of content distribution in the CDN environment, significantly improving the accuracy, adaptability, and practicality of predictions.

[0081] In some embodiments, the parameter training is achieved by constructing and minimizing the following objective function: the objective function includes the sum of errors between the predicted distribution amount and the actual historical distribution amount of the initial diffusion model on all training samples, and regularization constraints on the content push coefficient, the content request coefficient and the inter-node influence coefficient.

[0082] It should be noted that by constructing and optimizing this objective function, the model complexity can be effectively controlled while fitting historical data, preventing overfitting and improving the model's generalization ability and stability.

[0083] This application introduces a regularized objective function for model parameter training, which not only accurately fits historical distribution data but also effectively controls model complexity and improves the model's predictive stability and generalization ability on unseen data, thereby ensuring the reliability and practicality of content distribution volume prediction results in actual CDN operation.

[0084] This application provides an innovative CDN node distribution volume prediction method that applies the classic Bass diffusion theory to the CDN field. It can accurately and timely predict the distribution volume of content throughout its entire lifecycle (i.e., the entire period from the first release of the content to when it is no longer frequently accessed) at each level of the CDN. It is a prediction method with a theoretical basis, strong practicality and flexibility, and has the potential to significantly improve the content distribution efficiency and resource utilization of CDN.

[0085] Figure 3 This is a flowchart illustrating the CDN node content distribution volume prediction method provided in an embodiment of this application. Figure 3 As shown, the CDN node content distribution volume prediction method includes: Step 1: Data Preprocessing. Identify and classify influencing factors, collect corresponding historical data, and clean the historical data, removing outliers and missing values. Data preprocessing requires collecting and processing the following historical data: 1) Construct the training and testing datasets using historical distribution time-series data for each node. .

[0086] 2) Node characteristic data (such as node level, geographical location, bandwidth capacity, etc.), which will affect the network's maximum distribution capacity. .

[0087] 3) Content characteristic data (such as content type, size, publication time, etc.), these data will affect the content push coefficient. and content request coefficient The estimate.

[0088] 4) Data on external influencing factors (such as the impact of marketing campaigns, recommendation systems, etc.), for example, constructing content importance variables. Popularity variables .

[0089] 5) Internal influencing factor data (such as user ratings, sharing rates, etc.), for example, constructing content relevance variables. .

[0090] Step Two: Model Expansion. Based on preprocessed historical data and the experience of CDN experts, determine the additional variables that need to be introduced. These mainly affect the overall structure of the model and prepare for subsequent parameter estimation. The basic Bass model can be flexibly expanded according to the specific needs of CDN, such as adding external variables, like holiday effects or the impact of sudden events. This adaptability allows the model to more accurately capture the complex factors in the CDN environment.

[0091] Step 3: Parameter Estimation. Using methods such as nonlinear least squares, the model's parameters are fitted on the historical data training set. Compared to complex machine learning models, the parameter estimation process of the Bass model is computationally fast and resource-efficient. This makes it particularly suitable for real-time prediction and dynamic adjustment in CDN environments, enabling rapid response to changes in network conditions.

[0092] Step 4: Generate Distribution Volume Predictions. Based on the parameters fitted in Step 3 and the current state values ​​of each variable, substitute them into the extended Bass model determined in Step 2 to generate a future content distribution volume prediction curve. Generating prediction results is the core of model application. The model can not only produce content distribution volume predictions at different node granularities, but also generate model parameters for business insights. Furthermore, based on the prediction values, model parameters, and model structure, data insights such as S-curves and distribution peaks can be derived.

[0093] Step 5: Model Validation and Tuning. Using a validation set of historical data, evaluate the model's predictive ability by comparing predicted and actual values. Based on the validation results, adjust model parameters or structure, and return to Step 2 or Step 3 for iteration. The model validation and tuning process ensures the model's accuracy and reliability. On the historical data validation set, calculate the prediction error, such as root mean square error (RMSE) or mean absolute percentage error (MAPE), to evaluate the model's performance. Adjust model parameters or structure based on the evaluation results.

[0094] Step Six: Apply the Predictive Model. Apply the optimized predictive model to real-time online data to obtain continuously updated content distribution volume predictions. During the application of the predictive model, it is important to update the model parameters regularly (e.g., daily or weekly) with the latest data and continuously monitor the deviation between the predicted results and the actual distribution volume to prevent the model's effectiveness from declining.

[0095] Each step of the above content distribution volume prediction process is associated with different components of the Bass model, gradually building a content distribution volume prediction model suitable for CDN scenarios. The influence factors of both static historical data and dynamic real-time data on content distribution volume are calculated, enabling the prediction model to continuously adjust and optimize itself to adapt to the ever-changing CDN environment.

[0096] This application breaks through the current passive situation where content distribution scheduling relies entirely on passive scheduling based on user requests or active injection based on operational experience. It applies the Bass diffusion model to the field of CDN content distribution volume prediction, giving full play to the significant advantages of the model and organically combining theoretical foundation, practicality, and flexibility. It provides CDN content operators with a powerful and easy-to-use prediction tool, the advantages of which are reflected in: 1. Using a simple yet powerful algorithm model, it not only builds accurate content distribution prediction capabilities but also ensures the interpretability and flexibility of the prediction model, reserving room for continuous optimization.

[0097] 2. The parameters have clear business meanings, making them easy for content operators to understand and apply.

[0098] 3. It has strong long-term predictive capabilities and can capture distribution characteristics at different stages, making it particularly suitable for predicting the combined distribution volume of short-term hot content and long-tail content in CDN.

[0099] 4. Highly adaptable and easily expandable, it can be adjusted at any time according to the changing trends of CDN content.

[0100] 5. High computational efficiency, suitable for real-time applications and rapid response.

[0101] 6. It has a strong ability to handle sparse data and is suitable for predicting new content.

[0102] 7. Supports multi-scale prediction, providing comprehensive and in-depth insights for CDN content operations.

[0103] The content scheduling apparatus provided in the embodiments of this application is described below. The content scheduling apparatus described below and the content scheduling method described above can be referred to in correspondence.

[0104] Figure 4 This is a schematic diagram of the content scheduling device provided in an embodiment of this application. Figure 4 As shown, the content scheduling device 400 includes: The data acquisition module 410 is used to acquire attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and data of external influencing factors. The distribution volume prediction module 420 is used to input the attribute data of the content to be distributed, the status data of the at least one target node, and the external influencing factor data into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network. The content scheduling module 430 is used to pre-distribute content and / or adjust resources for the at least one target node based on the predicted distribution volume information.

[0105] In some embodiments, the data acquisition module 410 includes: The first acquisition unit is used to acquire at least one of the type, size and publication time of the content to be distributed as attribute data of the content to be distributed; The second acquisition unit is used to acquire at least one of the hierarchical position, geographical location and bandwidth capacity of the at least one target node as the status data of the at least one target node; The third acquisition unit is used to acquire at least one of the following as external influencing factor data: seasonal information, marketing activity information, recommendation system information, and social hot topic information.

[0106] In some embodiments, the distribution volume prediction module 420 is specifically used for: Output the predicted cumulative distribution volume curve of the content to be distributed on at least one target node over time; And / or, output the predicted peak distribution time and peak distribution amount of the content to be distributed on at least one target node.

[0107] In some embodiments, the content scheduling module 430 is specifically used for: When the peak distribution volume in the predicted distribution volume information exceeds a preset first threshold, a pre-distribution instruction is generated to actively push the content to be distributed from the upper-level node or the source station to the at least one target node. And / or, when the total predicted distribution amount for the target node in the predicted distribution amount information exceeds a preset second threshold, an adjustment suggestion for bandwidth expansion or load balancing of the at least one target node is generated.

[0108] In some embodiments, the apparatus further includes a model acquisition module, the model acquisition module comprising: The data collection unit is used to collect the historical operation dataset of the content distribution network. The historical operation dataset includes historical content attribute data of multiple historical contents, as well as historical distribution volume time-series data, historical node status data and historical external influencing factor data of each node. The model training unit is used to train the parameters of the initial diffusion model based on Bass diffusion theory using the historical operation dataset as training samples, so as to obtain the trained content distribution volume prediction model and its model parameter set.

[0109] In some embodiments, the first node is any node in the content delivery network, and the initial diffusion model models the distribution process of the first node as follows: The instantaneous distribution change rate of the first node at the target time is equal to the sum of the first and second terms multiplied by the remaining distribution capacity of the first node; The first term is the content push coefficient representing the proactive distribution capability of the first node itself, and the second term is the product of the content request coefficient representing the user request-driven capability of the first node and the current cumulative distribution volume ratio.

[0110] In some embodiments, the initial diffusion model also introduces an inter-node influence coefficient to characterize the probability of content spreading from other nodes to the first node; When modeling the first node, the sum of the first and second terms also includes a third term, which is the sum of the current cumulative distribution ratio of other nodes after being weighted by the influence coefficient between the corresponding nodes.

[0111] In some embodiments, the initial diffusion model further introduces at least one dynamic influence factor function, which is used to dynamically adjust the distribution process of the first node based on the input content attribute data, node status data and / or external influencing factor data.

[0112] In some embodiments, the dynamic impact factor function includes at least one of the following: Importance factor function, used to characterize the decay of content value over time; The relevance factor function is used to characterize the degree of matching between content and user groups; Seasonal factor function, used to characterize the fluctuation of distribution volume over time period; Popularity factor function is used to characterize the growth trend of content popularity.

[0113] In some embodiments, the parameter training is achieved by constructing and minimizing the following objective function: the objective function includes the sum of errors between the predicted distribution amount and the actual historical distribution amount of the initial diffusion model on all training samples, and regularization constraints on the content push coefficient, the content request coefficient and the inter-node influence coefficient.

[0114] Figure 5 This is a schematic diagram of the physical structure of the electronic device provided in the embodiments of this application. For example... Figure 5 As shown, the electronic device 500 may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communication interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 may call a computer program in the memory 530 to execute the steps of a content scheduling method, such as including: Acquire attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and data on external influencing factors; The attribute data of the content to be distributed, the status data of the at least one target node, and the data of the external influencing factors are input into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network. Based on the predicted distribution volume information, content pre-distribution and / or resource adjustment are performed on the at least one target node.

[0115] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] On the other hand, embodiments of this application also provide a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can perform the steps of the content scheduling method provided in the above embodiments, such as including: Acquire attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and data on external influencing factors; The attribute data of the content to be distributed, the status data of the at least one target node, and the data of the external influencing factors are input into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network. Based on the predicted distribution volume information, content pre-distribution and / or resource adjustment are performed on the at least one target node.

[0117] On the other hand, embodiments of this application also provide a processor-readable storage medium storing a computer program, the computer program being used to cause a processor to execute the steps of the content scheduling method provided in the above embodiments, for example including: Acquire attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and data on external influencing factors; The attribute data of the content to be distributed, the status data of the at least one target node, and the data of the external influencing factors are input into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network. Based on the predicted distribution volume information, content pre-distribution and / or resource adjustment are performed on the at least one target node.

[0118] The processor-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., CD, DVD, BD, HVD), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0119] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0120] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A content scheduling method, characterized in that, include: Acquire attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and data on external influencing factors; The attribute data of the content to be distributed, the status data of the at least one target node, and the data of the external influencing factors are input into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network. Based on the predicted distribution volume information, content pre-distribution and / or resource adjustment are performed on the at least one target node.

2. The content scheduling method according to claim 1, characterized in that, The acquisition of attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and current external influencing factor data includes: Obtain at least one of the type, size, and publication time of the content to be distributed as attribute data of the content to be distributed; Obtain at least one of the hierarchical position, geographical location, and bandwidth capacity of the at least one target node as the status data of the at least one target node; At least one of the following external influencing factors is obtained: seasonal information, marketing campaign information, recommendation system information, and social hot topic information.

3. The content scheduling method according to claim 1, characterized in that, The step of inputting the attribute data of the content to be distributed, the status data of the at least one target node, and the external influencing factor data into a trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node within a specified future time period includes: Output the predicted cumulative distribution volume curve of the content to be distributed on at least one target node over time; And / or, output the predicted peak distribution time and peak distribution amount of the content to be distributed on at least one target node.

4. The content scheduling method according to claim 1, characterized in that, The step of pre-distributing content or adjusting resources for at least one target node based on the predicted distribution volume information includes: When the peak distribution volume in the predicted distribution volume information exceeds a preset first threshold, a pre-distribution instruction is generated to actively push the content to be distributed from the upper-level node or the source station to the at least one target node. And / or, when the total predicted distribution amount for the target node in the predicted distribution amount information exceeds a preset second threshold, an adjustment suggestion for bandwidth expansion or load balancing of the at least one target node is generated.

5. The content scheduling method according to claim 1, characterized in that, The method further includes obtaining the trained content distribution volume prediction model, wherein obtaining the trained content distribution volume prediction model includes: Collect the historical operation dataset of the content distribution network. The historical operation dataset includes historical content attribute data of multiple historical contents, as well as historical distribution volume time-series data, historical node status data, and historical external influencing factor data of each node. Using the historical operation dataset as training samples, the parameters of the initial diffusion model built based on Bass diffusion theory are trained to obtain the trained content distribution volume prediction model and its model parameter set.

6. The content scheduling method according to claim 5, characterized in that, The first node is any node in the content delivery network, and the initial diffusion model models the distribution process of the first node as follows: The instantaneous distribution change rate of the first node at the target time is equal to the sum of the first and second terms multiplied by the remaining distribution capacity of the first node; The first term is the content push coefficient representing the proactive distribution capability of the first node itself, and the second term is the product of the content request coefficient representing the user request-driven capability of the first node and the current cumulative distribution volume ratio.

7. The content scheduling method according to claim 6, characterized in that, The initial diffusion model also introduces an inter-node influence coefficient to characterize the probability of content spreading from other nodes to the first node; When modeling the first node, the sum of the first and second terms also includes a third term, which is the sum of the current cumulative distribution ratio of other nodes after being weighted by the influence coefficient between the corresponding nodes.

8. The content scheduling method according to claim 6 or 7, characterized in that, The initial diffusion model also introduces at least one dynamic influence factor function, which is used to dynamically adjust the distribution process of the first node based on the input content attribute data, node status data and / or external influencing factor data.

9. The content scheduling method according to claim 8, characterized in that, The dynamic impact factor function includes at least one of the following: Importance factor function, used to characterize the decay of content value over time; The relevance factor function is used to characterize the degree of matching between content and user groups; Seasonal factor function, used to characterize the fluctuation of distribution volume over time period; Popularity factor function is used to characterize the growth trend of content popularity.

10. The content scheduling method according to claim 7, characterized in that, The parameter training is achieved by constructing and minimizing the following objective function: the objective function includes the sum of errors between the predicted distribution amount and the actual historical distribution amount of the initial diffusion model on all training samples, and regularization constraints on the content push coefficient, the content request coefficient and the inter-node influence coefficient.

11. A content scheduling device, characterized in that, include: The data acquisition module is used to acquire attribute data of the content to be distributed, status data of at least one target node in the content distribution network, and data on external influencing factors. The distribution volume prediction module is used to input the attribute data of the content to be distributed, the status data of the at least one target node, and the external influencing factor data into the trained content distribution volume prediction model to obtain the predicted distribution volume information of the content to be distributed on the at least one target node in a future specified time period. The trained content distribution volume prediction model is a diffusion model built based on the Bass diffusion model, which configures independent model parameter sets for different nodes in the content distribution network. The content scheduling module is used to pre-distribute content and / or adjust resources for the at least one target node based on the predicted distribution volume information.

12. An electronic device comprising a processor and a memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the content scheduling method according to any one of claims 1 to 10.

13. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the content scheduling method as described in any one of claims 1 to 10.

Images