Intelligent acceleration method and system for short play content distribution network
By constructing a hybrid request pattern graph of the short drama content distribution network, identifying and optimizing cache resource scheduling, the problem of distinguishing user request patterns is solved, and more efficient content pre-fetching and distribution are achieved.
Patent Information
- Application Number
- CN202512003405.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-29
- Publication Date
- 2026-01-27
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing technologies struggle to effectively distinguish and quantify the periodic patterns and sudden bursts of user requests in short-drama content distribution networks, leading to blind allocation of cache resources and impacting content prefetching hit rate and distribution efficiency.
By collecting historical request records from the edge nodes of the short drama content distribution network, identifying the mean, variance, skewness, and kurtosis of request intervals, constructing a hybrid request pattern graph, calculating connection strength and path complexity, generating an activity score, pruning the graph to form a core request pattern network, predicting future request traffic, and performing distributed preloading.
Accurately identify key behavioral patterns, optimize cache resource scheduling, improve content prefetch hit rate and distribution efficiency, and enhance user experience and service quality.
Smart Images
Figure CN121418596A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of content delivery network acceleration technology, specifically to an intelligent acceleration method and system for short drama content delivery networks. Background Technology
[0002] In short-drama content distribution networks, traditional intelligent acceleration and content prefetching strategies largely rely on holistic statistical analysis of historical request data. This approach typically treats user requests as independent or simply time-series events with simple temporal correlations, guiding caching decisions by calculating a global list of popular content. However, short-drama user viewing behavior exhibits a complex hybrid pattern, including both periodic, regular requests from following shows and sudden, intensive access triggered by social media sharing and platform recommendations. Existing technologies struggle to effectively separate and quantify the different impacts of these two patterns on the request graph and their inherent connections.
[0003] The current solution suffers from a coarse-grained modeling of user request patterns, lacking structured analysis and importance assessment of the behavioral patterns behind request sequences. When making predictions directly based on undifferentiated mixed historical data, the model is susceptible to noise patterns, resulting in insufficient insight into periodic patterns or inadequate preparation for sudden traffic surges. This leads to indiscriminate allocation of cache resources, potentially wasting bandwidth and storage resources on structurally unimportant or incidentally occurring request patterns, while truly critical user behavior paths are not prioritized, ultimately impacting content prefetching hit rates and overall distribution efficiency. Summary of the Invention
[0004] The purpose of this invention is to provide an intelligent acceleration method and system for short drama content distribution networks to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides an intelligent acceleration method for a short drama content distribution network. The method includes: collecting historical request records from multiple edge nodes of the short drama content distribution network, wherein the historical request records include a timestamp sequence of user requests and a corresponding short drama content identifier.
[0006] Multiple request interval sets are generated based on the interval between adjacent requests in the timestamp sequence. The mean, variance, skewness, and kurtosis of the time interval are calculated for each request interval set. Pattern recognition is performed on the request interval sets to extract periodic request patterns and burst request patterns.
[0007] By merging the periodic request pattern and the bursty request pattern, a hybrid request pattern map is constructed.
[0008] Calculate the connection strength and path complexity of each pattern node in the hybrid request pattern graph, generate an activity score for the pattern node based on the connection strength and path complexity, and prune the hybrid request pattern graph according to the activity score of the pattern node, retaining the pattern nodes and connection edges whose activity scores exceed the threshold, forming the core request pattern network.
[0009] Based on the core request pattern network prediction of request traffic distribution within a future time window, a prefetch instruction list is generated.
[0010] Based on the prefetch instruction list, cached resources in the short drama content distribution network are scheduled to perform distributed preloading of the short drama content.
[0011] Preferably, the step of performing pattern recognition on the request interval set to extract periodic request patterns and burst request patterns includes: for each request interval set, calculating the statistical characteristics of its time interval, wherein the statistical characteristics include mean, variance, skewness and kurtosis.
[0012] The statistical features are input into a pre-trained classification model, which outputs the probability that the request interval set belongs to a periodic pattern and the probability that it belongs to a burst pattern.
[0013] The set of request intervals with probability values exceeding the preset confidence level are classified into the corresponding periodic request pattern candidate set or burst request pattern candidate set.
[0014] For the candidate set of periodic request patterns, detect whether there is a stable repeating period in its time interval sequence, and mark the candidate set that confirms the existence of a stable repeating period as the final periodic request pattern.
[0015] For the candidate set of burst request patterns, detect whether there is a dense clustering of short intervals in its time interval sequence, and mark the candidate set that confirms the existence of the short interval clustering as the final burst request pattern.
[0016] Preferably, the step of merging the periodic request pattern and the burst request pattern to construct a hybrid request pattern graph includes: abstracting each finally marked periodic request pattern and burst request pattern into a pattern node.
[0017] If two pattern nodes have a sequential relationship in terms of time or a related relationship in terms of content in their historical request records, then a directed connection edge is established between the two pattern nodes.
[0018] Each directed connection edge is assigned a weight, which is calculated based on the frequency of the two pattern nodes appearing sequentially in the historical records and the average time interval between them.
[0019] All pattern nodes and weighted directed edges together constitute the hybrid request pattern graph.
[0020] Preferably, the calculation of the connection strength and path complexity of each pattern node in the hybrid request pattern graph includes: for any pattern node in the hybrid request pattern graph, the sum of the weights of all directed connection edges with the pattern node as the starting point or ending point is calculated as the connection strength of the pattern node.
[0021] Calculate the minimum number of edges required to reach any other reachable pattern node in the graph from the pattern node via directed connecting edges, and take the average of all such minimum edge numbers as the average path length of the pattern node.
[0022] Calculate the sum of the number of paths from the given pattern node to all other reachable pattern nodes, where each path does not repeatedly pass through the same pattern node. The product of the sum and the average path length is taken as the path complexity of the given pattern node.
[0023] Preferably, the step of generating the activity score of the pattern node based on connection strength and path complexity includes: normalizing the connection strength of all pattern nodes in the hybrid request pattern graph to obtain a normalized connection strength value for each pattern node.
[0024] The path complexity of all pattern nodes in the hybrid request pattern graph is normalized to obtain the normalized path complexity value of each pattern node.
[0025] The normalized connection strength value of the pattern node is multiplied by a weighted product of the normalized path complexity value of the pattern node minus one, and the product is then multiplied by a preset amplification factor. The result is the activity score of the pattern node.
[0026] Preferably, pruning the hybrid request pattern graph based on the activity scores of the pattern nodes includes: setting a dynamic activity score threshold, wherein the dynamic activity score threshold is determined based on the distribution of activity scores of all pattern nodes.
[0027] Traverse all pattern nodes in the hybrid request pattern graph and remove pattern nodes with activity scores below the dynamic threshold from the graph.
[0028] After removing a node, further remove all directed edges connected to the removed node.
[0029] Examine the subgraph formed by the remaining pattern nodes and directed connecting edges to ensure its connectivity, and use the final subgraph as the core request pattern network.
[0030] Preferably, the step of predicting the request traffic distribution within a future time window based on the core request pattern network includes: obtaining the current time point and matching the pattern node in the core request pattern network that is closest to the request characteristics of the current time point as the starting prediction node.
[0031] Starting from the initial prediction node, a request propagation process is simulated in the core request pattern network. The simulated request propagation process probabilistically selects the next pattern node based on the weights of the directed connection edges.
[0032] Record the sequence of pattern nodes accessed during the simulation, as well as the number of times each pattern node was accessed.
[0033] The simulation process is repeated multiple times, and the total frequency of each mode node being accessed in the multiple simulations is counted.
[0034] Based on the total frequency of access to each mode node and the intensity of the request mode it represents, the estimated request volume for short drama content associated with each mode node within a specific future time window is calculated, thereby generating the request traffic distribution.
[0035] Preferably, generating the prefetch instruction list includes: determining short drama content identifiers whose estimated request volume exceeds the prefetch threshold based on the request traffic distribution.
[0036] For each short drama content identifier, the optimal prefetch source node is calculated by combining the position of its corresponding mode node in the core request mode network and the current cache position of the short drama content in the content distribution network.
[0037] Generate a prefetch instruction, which includes the short drama content identifier, the corresponding optimal prefetch source node, and the prefetch priority determined based on the estimated request volume.
[0038] All generated prefetch instructions are sorted from high to low according to prefetch priority to form the prefetch instruction list.
[0039] Preferably, the step of scheduling cached resources in the short drama content distribution network to perform distributed preloading of short drama content includes: the central scheduling node in the short drama content distribution network receiving the prefetch instruction list.
[0040] The central scheduling node sends a content retrieval request to the corresponding source node based on the optimal prefetch source node information in the prefetch instruction list.
[0041] The source node responds to the content retrieval request by transmitting the specified short drama content in chunks to the edge node or intermediate cache node that issued the request.
[0042] During transmission, network bandwidth and cache space are dynamically allocated according to the prefetch priority in the prefetch instruction list to ensure that high-priority short drama content blocks are preloaded first.
[0043] Preferably, the present invention also includes an intelligent acceleration system for a short drama content distribution network. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. It also includes instructions for constructing a hybrid request pattern graph, calculating connection strength and path complexity, generating activity scores, performing graph pruning, and simulating the request propagation process. When the processor executes the computer program, it implements the steps of the intelligent acceleration method for the short drama content distribution network described above.
[0044] Compared with existing technologies, the advantages of this invention are: it quantifies and scores nodes in a hybrid request pattern graph based on "connection strength" and "path complexity," changing the traditional evaluation method that relies on a single frequency indicator. Connection strength measures the tightness of the association between a pattern node and other nodes, while path complexity characterizes the tortuousness of the behavioral sequence that triggers the pattern. Combining the two to calculate the activity score allows for a more accurate identification of patterns that not only appear frequently but also occupy a pivotal position in the global request flow or represent typical complex viewing paths, from both network structure and behavioral logic dimensions. This scoring mechanism enables the system to effectively distinguish between superficial high-frequency access and deep-seated key behavioral patterns.
[0045] Based on the aforementioned activity scores, the initial mixed request pattern graph is pruned, dynamically selecting high-scoring nodes and connecting edges to form the core request pattern network. This process essentially denoises and refines the behavioral knowledge extracted from the entire dataset, filtering out numerous accidental, secondary, or structurally insignificant patterns. Traffic prediction based on this core network has a cleaner and more focused input model, directly linking to key behavioral paths that impact user experience and service quality. This improves the accuracy of future request distribution predictions, making the generated prefetch instruction list more relevant to actual needs, thereby optimizing cache resource scheduling and achieving more precise distributed preloading. Attached Figure Description
[0046] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent acceleration method for short drama content distribution networks described in this invention.
[0047] Figure 2 This is a flowchart for calculating the connection strength and path complexity of pattern nodes.
[0048] Figure 3 A flowchart for pruning the graph to form a core request pattern network.
[0049] Figure 4Simulate the total frequency distribution of access for the mode node.
[0050] Figure 5 A comparison chart showing the success rate of short drama content transmission. Detailed Implementation
[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0052] Please see Figure 1 This invention provides an intelligent acceleration method for a short drama content distribution network. The method includes: analyzing patterns in historical request records to optimize content prefetching and cache scheduling. The method collects historical request records from multiple edge nodes, which contain timestamp sequences of user requests and corresponding short drama content identifiers. The interval between adjacent requests in the timestamp sequences is used to generate multiple request interval sets. Pattern recognition is performed on the request interval sets to extract periodic request patterns and bursty request patterns. Periodic and bursty request patterns are merged to construct a hybrid request pattern graph. The connection strength and path complexity of each pattern node in the hybrid request pattern graph are calculated, and an activity score for the pattern node is generated based on the connection strength and path complexity. The hybrid request pattern graph is pruned according to the activity score of the pattern node, retaining pattern nodes and connecting edges with activity scores exceeding a threshold to form a core request pattern network. The request traffic distribution within a future time window is predicted based on the core request pattern network, and a prefetch instruction list is generated. Cache resources in the short drama content distribution network are scheduled based on the prefetch instruction list to perform distributed preloading of short drama content.
[0053] Example 1: In the pattern recognition process, statistical features of the time intervals are calculated for each request interval set. These features include mean, variance, skewness, and kurtosis. The statistical features are input into a pre-trained classification model. The model outputs the probability that the request interval set belongs to a periodic pattern and the probability that it belongs to a burst pattern. Request interval sets with probability values exceeding a preset confidence level are classified into the corresponding periodic request pattern candidate set or burst request pattern candidate set. For the periodic request pattern candidate set, the presence of stable repetitive cycles in its time interval sequence is checked, and candidate sets with confirmed stable repetitive cycles are marked as the final periodic request pattern. For the burst request pattern candidate set, the presence of dense short-interval clusters in its time interval sequence is checked, and candidate sets with confirmed short-interval clusters are marked as the final burst request pattern. When constructing a hybrid request pattern graph, each finally labeled periodic and burst request pattern is abstracted into a pattern node. If two pattern nodes have a temporal sequence relationship or a content-related request relationship in their historical request records, a directed connection edge is established between the two pattern nodes. Each directed connection edge is assigned a weight, which is calculated based on the frequency of the two pattern nodes appearing sequentially in the historical records and the average time interval between their acceptance. All pattern nodes and weighted directed connections together constitute the hybrid request pattern graph.
[0054] In its implementation, the short drama content distribution network collects historical request records from multiple edge nodes. These records contain timestamp sequences of user requests and corresponding short drama content identifiers. Multiple request interval sets are generated based on the interval between adjacent requests in the timestamp sequence. For each request interval set, statistical characteristics of its time intervals are calculated, including mean, variance, skewness, and kurtosis. These characteristics are input into a pre-trained classification model. The model outputs the probability that each request interval set belongs to a periodic pattern and the probability that it belongs to a bursty pattern. Request interval sets with probability values exceeding a pre-set confidence level are classified into the corresponding periodic request pattern candidate set or bursty request pattern candidate set. Furthermore, for the periodic request pattern candidate set, the presence of stable repetitive cycles in its time interval sequence is detected, and candidate sets with confirmed stable repetitive cycles are marked as the final periodic request pattern. Similarly, for the bursty request pattern candidate set, the presence of dense short-interval clusters in its time interval sequence is detected, and candidate sets with confirmed dense short-interval clusters are marked as the final bursty request pattern.
[0055] In some embodiments, the periodic and bursty request patterns of each final tag are abstracted into a pattern node. If two pattern nodes have a temporal sequence relationship or a content-related request relationship in the historical request record, a directed connection edge is established between the two pattern nodes. In some embodiments, each directed connection edge is assigned a weight, which is calculated based on the frequency of the two pattern nodes appearing sequentially in the historical record and the average time interval between their occurrences. All pattern nodes and weighted directed connections edge by edge together constitute a hybrid request pattern graph. It can be understood that the weight calculation of the directed connection edge can be explicitly expressed using the following formula:
[0056] Where: symbol Indicates from the pattern node To the pattern node The weight of the directed connection edge, sign Represents the pattern node With pattern node Frequency of occurrence in historical records, symbols Represents the pattern node With pattern node The average time interval carried over in historical records, with the sign... It is a small positive number used to prevent the denominator from being zero and to smooth the weight calculation.
[0057] Optionally, the pre-trained classification model can be trained based on historically labeled request interval sets to distinguish the feature differences between periodic and burst patterns. Optionally, detecting stable recurring periods can be achieved by calculating the autocorrelation function or Fourier transform method of the time interval sequence.
[0058] Example 2: See Figure 2When calculating connection strength and path complexity, for any pattern node in the hybrid request pattern graph, the sum of the weights of all directed edges originating from or ending at that pattern node is used as the connection strength of the pattern node. The minimum number of edges required to reach any other reachable pattern node in the graph from that pattern node is calculated, and the average of all such minimum edge counts is used as the average path length of the pattern node. The total number of paths from that pattern node to all other reachable pattern nodes is calculated, where each path does not repeatedly pass through the same pattern node. The product of this total and the average path length is used as the path complexity of the pattern node. When generating activity scores, the connection strength of all pattern nodes in the hybrid request pattern graph is normalized to obtain a normalized connection strength value for each pattern node, and the path complexity of all pattern nodes in the hybrid request pattern graph is normalized to obtain a normalized path complexity value for each pattern node. The normalized connection strength value of the pattern node is multiplied by a weighted factor equal to the normalized path complexity value of the pattern node, and the resulting product is then multiplied by a preset amplification factor. The result is the activity score of the pattern node.
[0059] In practical implementation, when calculating the connection strength of each pattern node in the hybrid request pattern graph, for any pattern node in the graph, the sum of the weights of all directed edges originating from or ending at this pattern node is counted, and this sum is used as the connection strength of the pattern node. In practical implementation, when calculating the path complexity of a pattern node, it is necessary to calculate the minimum number of edges required to reach any other reachable pattern node in the graph from this pattern node via directed edges, and the average of all such minimum edge counts is used as the average path length of the pattern node. Then, the sum of the number of paths from this pattern node to all other reachable pattern nodes is calculated, and the product of this sum and the average path length is used as the path complexity of the pattern node.
[0060] In some embodiments, when generating activity scores for pattern nodes based on connection strength and path complexity, the connection strength of all pattern nodes in the hybrid request pattern graph is normalized to obtain a normalized connection strength value for each pattern node. In some embodiments, the path complexity of all pattern nodes in the hybrid request pattern graph is normalized to obtain a normalized path complexity value for each pattern node. It can be understood that the normalization process can employ a min-max normalization method, linearly transforming the original values to between zero and one.
[0061] The activity score of the pattern node is calculated using the following formula:
[0062] Where: symbol Represents the pattern node Activity rating, symbol Indicates the preset magnification factor, symbol Represents the pattern node The normalized connection strength value, sign Represents the pattern node The normalized path complexity value.
[0063] Optionally, the connection strength calculation can consider only the sum of the weights of outgoing edges originating from the pattern node, or it can consider the sum of the weights of both incoming and outgoing edges. Optionally, the total number of paths can be counted using a graph traversal algorithm, such as depth-first search or breadth-first search.
[0064] Example 3: See Figure 3 When pruning the hybrid request pattern graph, a dynamic activity score threshold is set, which is determined based on the distribution of activity scores across all pattern nodes. All pattern nodes in the hybrid request pattern graph are traversed, and those with activity scores below the dynamic threshold are removed from the graph. After removal, all directed edges connected to the removed node are further removed. The subgraph formed by the remaining pattern nodes and directed edges is checked and its connectivity is ensured. The final subgraph is used as the core request pattern network.
[0065] In practice, when pruning the hybrid request pattern graph based on the activity scores of pattern nodes, a dynamic activity score threshold is set. This threshold is determined based on the distribution of activity scores across all pattern nodes. All pattern nodes in the hybrid request pattern graph are traversed, and those with activity scores below the dynamic threshold are removed. After removal, all directed edges connecting these nodes are further removed. The subgraph formed by the remaining pattern nodes and their directed edges is checked for connectivity, and this subgraph is used as the core request pattern network.
[0066] In some embodiments, the dynamic activity score threshold can be set by analyzing the statistical characteristics of the activity scores of all mode nodes. In some embodiments, a specific method for setting the dynamic activity score threshold can be calculated using the following formula:
[0067] Where: symbol This represents the final set dynamic activity score threshold, symbol... This represents the arithmetic mean of the activity scores of all pattern nodes, with the sign... The standard deviation of the activity scores for all pattern nodes is represented by the symbol. It is an adjustable parameter factor used to control the degree of deviation of the threshold from the average value.
[0068] Optionally, after removing pattern nodes whose activity scores are below the dynamic activity score threshold, the connectivity of the remaining subgraph can be checked using a depth-first search or disjoint-set data structure algorithm in graph theory. Optionally, the parameter factor " The value of "" can be based on historical data or empirical presets to finely control the sparsity of the core request pattern network. It can be understood that checking and ensuring the connectivity of the subgraph is to prevent the graph from splitting into multiple unconnected isolated parts after pruning.
[0069] Example 4: When predicting request traffic distribution within a future time window, the current time point is obtained, and the pattern node with the closest request characteristics to the current time point is matched in the core request pattern network as the starting prediction node. Starting from the starting prediction node, the request propagation process is simulated in the core request pattern network. The simulation process probabilistically selects the next pattern node based on the weights of the directed connection edges. The sequence of pattern nodes accessed during the simulation and the number of times each pattern node is accessed are recorded. The simulation process is repeated multiple times, and the total frequency of access to each pattern node in multiple simulations is counted. Based on the total frequency of access to each pattern node and the strength of the request pattern it represents, the estimated request volume for short drama content associated with each pattern node within a specific future time window is estimated, thereby generating a request traffic distribution. When generating the prefetch instruction list, the short drama content identifiers whose estimated request volume exceeds the prefetch threshold are determined based on the request traffic distribution. For each short drama content identifier, the optimal prefetch source node is calculated by combining the position of its corresponding pattern node in the core request pattern network and the current cache position of the short drama content in the content distribution network. Generate a prefetch instruction, which includes the short drama content identifier, the corresponding optimal prefetch source node, and the prefetch priority determined based on the estimated request volume. Sort all generated prefetch instructions from high to low according to their prefetch priority to form a prefetch instruction list.
[0070] In practical implementation, when predicting the request traffic distribution within a future time window based on the core request pattern network, the current time point is obtained, and the pattern node with the closest request characteristics to the current time point is matched in the core request pattern network. The matched pattern node is used as the starting prediction node. Starting from the starting prediction node, the request propagation process is simulated in the core request pattern network. The simulation process probabilistically selects the next pattern node based on the weights of the directed connection edges, and records the sequence of pattern nodes visited during the simulation process and the number of times each pattern node is visited. The simulation process is repeated multiple times, and the total frequency of each pattern node being visited in multiple simulations is counted. Based on the total frequency of each pattern node being visited and the strength of the request pattern it represents, the estimated request volume for short drama content associated with each pattern node within a specific future time window is estimated, thereby generating the request traffic distribution. In some embodiments, the strength of the request pattern can be quantified by the size of the set of request intervals represented by the pattern node in the historical record or the total number of requests. In some embodiments, the simulated request propagation process can be set with a fixed number of simulation steps or a termination condition, such as stopping when the accumulated weight of the simulated path falls below a specific threshold.
[0071] A formula for estimating the number of requests for short drama content is expressed as follows:
[0072] Where: symbol This indicates the relationship between the pattern node and the time window in the future. Estimated request volume for related short drama content, symbol Represents the pattern node The total frequency accessed in multiple simulations, symbol Represents the pattern node The strength of the request pattern represented, symbol This indicates the length of a specific future time window. It can be understood that this formula estimates the request volume based on the product of access frequency and pattern strength, taking into account the impact of the time window. See Table 1 for the statistical results of the simulated access frequency of the pattern nodes.
[0073] In practice, when generating the prefetch instruction list, the short drama content identifiers whose estimated request volume exceeds the prefetch threshold are determined based on the request traffic distribution. For each short drama content identifier, the optimal prefetch source node is calculated by combining its corresponding mode node's position in the core request mode network and the short drama content's current cache position in the content distribution network. A prefetch instruction is generated, containing the short drama content identifier, the corresponding optimal prefetch source node, and a prefetch priority determined based on the estimated request volume. All generated prefetch instructions are sorted from high to low prefetch priority to form a prefetch instruction list. It can be understood that the prefetch threshold is used to filter out content with low estimated request volume to avoid invalid prefetch operations. Optionally, calculating the optimal prefetch source node can comprehensively consider network latency, link bandwidth, and source node load. Optionally, the determination of prefetch priority can be directly related to the estimated request volume. "It is directly proportional, meaning that the higher the estimated request volume of short drama content, the higher its prefetching priority."
[0074] See Figure 4 In predicting request traffic distribution in the core request pattern network, the simulated total access frequency of pattern nodes is the core foundational data for estimating the expected request volume of short drama content. Specifically, this chart quantifies the total access frequency of each node during multiple request propagation simulations, using the pattern node identifiers (P001 to P008) as the dimension. Among them, node P003 has a simulated total access frequency of 203, making it the node with the highest access popularity in the current core request pattern network; node P007 has a total access frequency of 76, making it the node with the lowest popularity. This frequency data, combined with the request pattern intensity and future time window length of the corresponding pattern node, is used to generate the estimated request volume of the short drama content associated with each node through a preset calculation formula, thereby supporting the priority ranking of the prefetch instruction list and the scheduling of cache resources.
[0075] Example 5: When scheduling cache resources for distributed preloading, the central scheduling node in the short drama content distribution network receives a prefetch instruction list. Based on the optimal prefetch source node information in the list, the central scheduling node sends a content retrieval request to the corresponding source node. The source node responds to the content retrieval request by transmitting the specified short drama content in chunks to the requesting edge node or intermediate cache node. During transmission, network bandwidth and cache space are dynamically allocated according to the prefetch priority in the prefetch instruction list to ensure that high-priority short drama content chunks are preloaded first.
[0076] In practical implementation, when scheduling cached resources in the short drama content distribution network and performing distributed preloading of short drama content, the central scheduling node in the short drama content distribution network receives a prefetch instruction list. The central scheduling node then sends a content retrieval request to the corresponding source node based on the optimal prefetch source node information in the prefetch instruction list. The source node responds to the content retrieval request and transmits the specified short drama content in chunks to the edge node or intermediate cache node that issued the request. During transmission, network bandwidth and cache space are dynamically allocated according to the prefetch priority in the prefetch instruction list to ensure that high-priority short drama content chunks are preloaded first. In some embodiments, the content retrieval request can be initiated in parallel or serial mode, depending on the central scheduling node's scheduling strategy and the current network load. In some embodiments, dynamic allocation of network bandwidth can be achieved by adjusting the data stream weights or priority queues of different prefetching tasks.
[0077] It is understandable that the optimal prefetch source node information in the prefetch instruction list includes the network location identifier of the target short drama content block. It is also understandable that dynamically allocating cache space can pre-reserve storage areas of appropriate size on edge nodes or intermediate cache nodes based on prefetch priority. The strategy for the central scheduling node to allocate transmission resources for each prefetch instruction can be expressed by the following formula:
[0078] Where: symbol Indicates that it is assigned to the identifier. The network bandwidth quota corresponding to the short drama content block of the prefetch instruction, symbol Indicates that the identifier is The prefetch priority value and sign of the prefetch instruction. This represents the sum of the prefetch priority values of all N prefetch tasks currently in the transmission state, with the sign... This represents the total network bandwidth currently available for prefetching operations.
[0079] Optionally, before sending a content retrieval request, the central scheduling node can first establish a connection with the optimal prefetch source node and verify its availability and load status. Optionally, the segmented transmission of short drama content can adopt a breakpoint resume mechanism to cope with abnormal interruptions that may occur during network transmission.
[0080] See Figure 5In the performance evaluation of the short drama content distribution network, the success rate variation characteristics during multiple transmission rounds were presented under two mechanisms: normal transmission and interrupted resume transmission. Specifically, the transmission success rate of the interrupted resume mechanism remained in a relatively high range of 83%-98%, and showed a gradual downward trend with the increase of transmission rounds; while the success rate of normal transmission rapidly decreased from an initial 98%, reaching only 70% by the 10th transmission round. The "success rate improvement" range between the two (the light-colored area in the figure) intuitively reflects the optimization effect of interrupted resume transmission on transmission stability. This mechanism effectively reduces the success rate loss in multiple transmission rounds through chunked transmission and interrupt recovery capabilities, which is highly consistent with the transmission reliability requirements of distributed preloading in the short drama content distribution network.
[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0082] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. An intelligent acceleration method for short drama content distribution networks, characterized in that, The method includes: Historical request records are collected from multiple edge nodes of the short drama content distribution network. These historical request records include a sequence of timestamps of user requests and the corresponding short drama content identifiers. Multiple request interval sets are generated based on the interval between adjacent requests in the timestamp sequence. The mean, variance, skewness, and kurtosis of the time interval are calculated for each request interval set. Pattern recognition is performed on the request interval sets to extract periodic request patterns and burst request patterns. By merging the periodic request pattern and the bursty request pattern, a hybrid request pattern graph is constructed; Calculate the connection strength and path complexity of each pattern node in the hybrid request pattern graph, generate an activity score for the pattern node based on the connection strength and path complexity, and prune the hybrid request pattern graph according to the activity score of the pattern node, retaining the pattern nodes and connection edges whose activity scores exceed the threshold, to form the core request pattern network. Based on the core request pattern network prediction of request traffic distribution within a future time window, a prefetch instruction list is generated; Based on the prefetch instruction list, cached resources in the short drama content distribution network are scheduled to perform distributed preloading of the short drama content.
2. The intelligent acceleration method for short drama content distribution networks according to claim 1, characterized in that, The step of performing pattern recognition on the request interval set to extract periodic request patterns and bursty request patterns includes: For each set of request intervals, calculate the statistical characteristics of its time intervals, including mean, variance, skewness, and kurtosis. The statistical features are input into a pre-trained classification model, which outputs the probability that the request interval set belongs to a periodic pattern and the probability that it belongs to a burst pattern. The set of request intervals with probability values exceeding the preset confidence level are classified into the corresponding periodic request pattern candidate set or burst request pattern candidate set. For the candidate set of periodic request patterns, detect whether there is a stable repeating period in its time interval sequence, and mark the candidate set that is confirmed to have a stable repeating period as the final periodic request pattern. For the candidate set of burst request patterns, detect whether there is a dense clustering of short intervals in its time interval sequence, and mark the candidate set that confirms the existence of the short interval clustering as the final burst request pattern.
3. The intelligent acceleration method for short drama content distribution networks according to claim 2, characterized in that, The process of merging the periodic request pattern and the bursty request pattern to construct a hybrid request pattern graph includes: Abstract the periodic request pattern and bursty request pattern of each final tag into a pattern node; If two pattern nodes have a sequential relationship in terms of time or a related relationship in terms of content in their historical request records, then a directed connection edge is established between the two pattern nodes. Each directed connection edge is assigned a weight, which is calculated based on the frequency of the two pattern nodes appearing sequentially in the historical records and the average time interval between them. All pattern nodes and weighted directed edges together constitute the hybrid request pattern graph.
4. The intelligent acceleration method for short drama content distribution networks according to claim 3, characterized in that, The calculation of the connection strength and path complexity of each pattern node in the hybrid request pattern graph includes: For any pattern node in the hybrid request pattern graph, the sum of the weights of all directed connection edges starting from or ending at the pattern node is counted as the connection strength of the pattern node. Calculate the minimum number of edges required to reach any other reachable pattern node in the graph from the pattern node via directed connecting edges, and take the average of all such minimum edge numbers as the average path length of the pattern node. Calculate the sum of the number of paths from the given pattern node to all other reachable pattern nodes, where each path does not repeatedly pass through the same pattern node. The product of the sum and the average path length is taken as the path complexity of the given pattern node.
5. The intelligent acceleration method for short drama content distribution networks according to claim 4, characterized in that, The activity score for generating pattern nodes based on connection strength and path complexity includes: The connection strength of all pattern nodes in the hybrid request pattern graph is normalized to obtain the normalized connection strength value of each pattern node. The path complexity of all pattern nodes in the hybrid request pattern graph is normalized to obtain the normalized path complexity value of each pattern node. The normalized connection strength value of the pattern node is multiplied by a weighted product of the normalized path complexity value of the pattern node minus one, and the product is then multiplied by a preset amplification factor. The result is the activity score of the pattern node.
6. The intelligent acceleration method for short drama content distribution networks according to claim 5, characterized in that, The step of pruning the hybrid request pattern graph based on the activity score of the pattern nodes includes: A dynamic activity score threshold is set, which is determined based on the distribution of activity scores of all mode nodes; Traverse all pattern nodes in the hybrid request pattern graph and remove pattern nodes with activity scores lower than the dynamic threshold from the graph; After removing a node, further remove all directed edges connected to the removed node; Examine the subgraph formed by the remaining pattern nodes and directed connecting edges to ensure its connectivity, and use the final subgraph as the core request pattern network.
7. The intelligent acceleration method for short drama content distribution networks according to claim 6, characterized in that, The method of predicting request traffic distribution within a future time window based on the core request pattern network includes: Obtain the current time point and match the pattern node in the core request pattern network that is closest to the request features at the current time point as the starting prediction node; Starting from the initial prediction node, a request propagation process is simulated in the core request pattern network. The simulated request propagation process probabilistically selects the next pattern node based on the weight of the directed connection edge. Record the sequence of pattern nodes accessed during the simulation, as well as the number of times each pattern node was accessed; The simulation process is repeated multiple times, and the total frequency of each mode node being accessed in the multiple simulations is counted. Based on the total frequency of access to each mode node and the intensity of the request mode it represents, the estimated request volume for short drama content associated with each mode node within a specific future time window is calculated, thereby generating the request traffic distribution.
8. The intelligent acceleration method for short drama content distribution networks according to claim 7, characterized in that, The generated prefetch instruction list includes: Based on the request traffic distribution, identify short drama content whose estimated request volume exceeds the prefetch threshold; For each short drama content identifier, the optimal prefetch source node is calculated by combining the position of its corresponding mode node in the core request mode network and the current cache position of the short drama content in the content distribution network. Generate a prefetch instruction, which includes the short drama content identifier, the corresponding optimal prefetch source node, and the prefetch priority determined based on the estimated request volume; All generated prefetch instructions are sorted from high to low according to prefetch priority to form the prefetch instruction list.
9. The intelligent acceleration method for short drama content distribution networks according to claim 8, characterized in that, The process of scheduling cached resources in the short drama content distribution network to perform distributed preloading of short drama content includes: The central scheduling node in the short drama content distribution network receives the prefetch instruction list; The central scheduling node sends a content retrieval request to the corresponding source node based on the optimal prefetch source node information in the prefetch instruction list. The source node responds to the content retrieval request by transmitting the specified short drama content in chunks to the edge node or intermediate cache node that issued the request. During transmission, network bandwidth and cache space are dynamically allocated according to the prefetch priority in the prefetch instruction list to ensure that high-priority short drama content blocks are preloaded first.
10. An intelligent acceleration system for a short drama content distribution network, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, further comprising instructions for constructing a hybrid request pattern graph, calculating connection strength and path complexity, generating activity scores, performing graph pruning, and simulating the request propagation process, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent acceleration method for the short drama content distribution network as described in any one of claims 1 to 9.