Data pushing method and device

Abstract

The embodiment of the invention provides a data pushing method and device. The method comprises the steps that response time data and throughput rate data of a data pushing interface of a user side are collected; the method comprises the following steps: decomposing reaction time data and throughput rate data into a trend part and a residual part based on a time sequence decomposition technology, respectively performing multi-dimensional feature extraction on the trend part and the residual part, and performing feature combination on trend features and residual features; performing feature selection on the feature vectors after feature combination according to a preset recursive feature elimination algorithm, determining interface performance features, training an initial model based on the interface performance features, and determining a corresponding interface performance prediction model; according to the preset time interval and the interface performance prediction model, performing secondary fine-grained performance prediction on the user side interface, determining the corresponding message push queue according to the prediction result and the preset matching rule, and pushing the push data to the user side through the message push queue, and the data push efficiency and flexibility of the push platform can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing, and particularly to a data push method and apparatus. Background Art

[0002] In existing receipt push platforms, there are significant differences in the performance of the receipt receiving interfaces of each customer. The traditional receipt push method usually places the receipt data of all customers in a unified queue for processing. This processing method has the following problems and disadvantages:

[0003] Mutual performance influence: Due to the inconsistent performance of the receipt interfaces of each customer, the customer interfaces with lower performance will slow down the processing speed of the entire receipt push platform, thereby affecting the receipt push efficiency of other customers with higher performance.

[0004] Data backlog: When the receipt push speed cannot meet the needs of all customers, the receipt data will backlog in the push platform, resulting in delays and lags in data processing.

[0005] Deteriorated customer experience: The excessive receipt response time will lead to an increase in the customer complaint rate, affecting customer satisfaction and loyalty. Especially in industries such as e-commerce and finance that require efficient and real-time responses, the slow receipt push speed will directly affect the normal operation of the business.

[0006] Resource waste: For customer interfaces with lower performance, the platform needs to invest more resources (such as computing resources, storage resources, etc.) to ensure the push of their receipt data. This not only increases the operating cost of the platform but also may lead to waste of server resources.

[0007] To address the above problems, existing receipt push platforms lack an effective mechanism to dynamically adjust the interface push levels of different customers to adapt to the different performance requirements of each customer interface. Therefore, there is an urgent need for a data push method that can improve the efficiency and flexibility of data push on the push platform. Summary of the Invention

[0008] To solve the problems in the prior art, this application provides a data push method and apparatus, which can improve the efficiency and flexibility of data push on the push platform.

[0009] To solve at least one of the above problems, this application provides the following technical solutions:

[0010] In a first aspect, this application provides a data push method, including:

[0011] Collecting the response time data of the user-side data push interface according to the time bucket aggregation technology, and collecting the throughput data of the user-side data push interface according to the sliding window technology;

[0012] Decompose the reaction time data and the throughput data into a trend part and a residual part based on time series decomposition technology, perform multi-dimensional feature extraction on the trend part and the residual part respectively, determine the corresponding trend features and residual features, combine the trend features and the residual features, perform feature selection on the feature vector after the feature combination according to a preset recursive feature elimination algorithm, determine the corresponding interface performance features, and train an initial model based on the interface performance features to determine the corresponding interface performance prediction model, where the trend features represent long-term change trend features, and the residual features represent random fluctuation features other than the trend features;

[0013] Perform secondary fine-grained performance prediction on the client interface according to a preset time interval and the interface performance prediction model, determine the corresponding first interface performance prediction data and second interface performance prediction data, determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, determine the corresponding message push queue according to the first-level queue and the second-level queue, and push the push data to the client through the message push queue.

[0014] Further, the collection of the reaction time data of the client data push interface according to the time binning aggregation technology includes:

[0015] Divide the time into multiple buckets of a fixed size;

[0016] Perform aggregation calculation on the reaction time of the client data push interface in each bucket to determine the corresponding reaction time data.

[0017] Further, the decomposition of the reaction time data and the throughput data into a trend part and a residual part based on time series decomposition technology includes:

[0018] Perform time series data decomposition on the reaction time data based on the Hodrick-Prescott filtering technology to determine the corresponding reaction time trend and reaction time residual;

[0019] Perform time series data decomposition on the throughput data based on the Hodrick-Prescott filtering technology to determine the corresponding throughput trend and throughput residual;

[0020] Determine the corresponding trend part according to the reaction time trend and the throughput trend, and determine the corresponding residual part according to the reaction time residual and the throughput residual.

[0021] Further, the multi-dimensional feature extraction of the trend part and the residual part respectively to determine the corresponding trend features and residual features includes:

[0022] Perform multi-dimensional feature extraction on the trend part using polynomial fitting to determine the corresponding trend features, where the trend features include trend slope, trend volatility, and trend extreme values;

[0023] Perform multi-dimensional feature extraction on the residual part using frequency domain analysis methods to determine the corresponding residual features, where the residual features include residual volatility, residual autocorrelation, and residual extreme values.

[0024] Further, the training of the initial model based on the interface performance features to determine the corresponding interface performance prediction model includes:

[0025] Construct corresponding performance index labels based on the secondary fine-grained performance indicators;

[0026] Input the interface performance features and the performance index labels into the initial model for model training to determine the corresponding interface performance prediction model, and the interface performance prediction model is used for secondary fine-grained performance prediction.

[0027] Further, the secondary fine-grained performance prediction of the user-side interface according to the preset time interval and the interface performance prediction model to determine the corresponding first interface performance prediction data and second interface performance prediction data, and determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, and determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, includes:

[0028] Perform secondary fine-grained performance prediction of the user-side interface according to the preset time interval and the interface performance prediction model to determine the corresponding first interface performance prediction data and second interface performance prediction data, where the first interface performance prediction data is used to represent the coarse-grained and fine-grained interface performance indicators, and the second interface performance prediction data is used to represent the fine-grained and fine-grained interface performance indicators;

[0029] Determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, and determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, where the first-level queue and the second-level queue are nested structures.

[0030] Further, the determination of the corresponding message push queue according to the first-level queue and the second-level queue includes:

[0031] Determine the overall priority of message push corresponding to the first-level queue, and determine the fine-grained priority of message push corresponding to the second-level queue;

[0032] Determine the corresponding message push queue according to the overall priority of message push and the fine-grained priority of message push.

[0033] In a second aspect, the present application provides a data push device, including:

[0034] A data collection module, configured to collect response time data of the user-side data push interface according to the time bucketing aggregation technology, and collect throughput data of the user-side data push interface according to the sliding window technology;

[0035] A machine learning module, configured to decompose the response time data and the throughput data into a trend part and a residual part based on the time series decomposition technology, perform multi-dimensional feature extraction on the trend part and the residual part respectively, determine the corresponding trend features and residual features, combine the trend features and the residual features, perform feature selection on the feature vector after the feature combination according to the preset recursive feature elimination algorithm, determine the corresponding interface performance features, and train an initial model based on the interface performance features to determine the corresponding interface performance prediction model, where the trend features represent long-term change trend features, and the residual features represent random fluctuation features other than the trend features;

[0036] A queue management module, configured to perform secondary fine-grained performance prediction on the user-side interface according to a preset time interval and the interface performance prediction model, determine the corresponding first interface performance prediction data and second interface performance prediction data, determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, determine the corresponding message push queue according to the first-level queue and the second-level queue, and push the push data to the user-side through the message push queue.

[0037] In a third aspect, the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the steps of the data push method described above are implemented.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the data push method described above are implemented.

[0039] Fifth aspect, the present application provides a computer program product, including a computer program / instructions, which when executed by a processor implement the steps of the data push method described above.

[0040] As can be seen from the above technical solutions, the present application provides a data push method and device. By collecting the response time data and throughput data of the user-side data push interface; decomposing the response time data and throughput data into a trend part and a residual part based on time series decomposition technology, performing multi-dimensional feature extraction on the trend part and the residual part respectively, combining the trend features and the residual features, performing feature selection on the feature vector after feature combination according to a preset recursive feature elimination algorithm to determine the interface performance features, training an initial model based on the interface performance features to determine the corresponding interface performance prediction model; performing secondary fine-grained performance prediction on the user-side interface according to a preset time interval and the interface performance prediction model, determining the corresponding message push queue according to the prediction result and a preset matching rule, and pushing the push data to the user-side through the message push queue, thereby improving the efficiency and flexibility of data push on the push platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required to be used in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1 is one of the flow diagrams of the data push method in the embodiments of the present application;

[0043] Figure 2 is another flow diagram of the data push method in the embodiments of the present application;

[0044] Figure 3 is yet another flow diagram of the data push method in the embodiments of the present application;

[0045] Figure 4 is still another flow diagram of the data push method in the embodiments of the present application;

[0046] Figure 5 is yet again a flow diagram of the data push method in the embodiments of the present application;

[0047] Figure 6 is still a flow diagram of the data push method in the embodiments of the present application;

[0048] Figure 7 is a further flow diagram of the data push method in the embodiments of the present application;

[0049] Figure 8 The structural diagram of the data push device in the embodiments of the present application;

[0050] Figure 9 The structural schematic diagram of the electronic device in the embodiments of the present application.

[0051] Reference numerals:

[0052] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver program storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed implementation manners

[0053] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Apparently, the described embodiments are some but not all of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0054] In the technical solutions of the present application, the acquisition, storage, use, processing, etc. of data all comply with the relevant provisions of national laws and regulations.

[0055] Considering the problem that the existing receipt push platform lacks an effective mechanism to dynamically adjust the interface push levels of different customers to meet the different performance requirements of each customer's interface. The present application provides a data push method and device, which collect the response time data and throughput data of the user-side data push interface; decompose the response time data and throughput data into a trend part and a residual part based on time series decomposition technology, perform multi-dimensional feature extraction on the trend part and the residual part respectively, combine the trend features and the residual features, perform feature selection on the feature vector after feature combination according to a preset recursive feature elimination algorithm to determine the interface performance features, train an initial model based on the interface performance features to determine the corresponding interface performance prediction model; perform secondary fine-grained performance prediction on the user-side interface according to a preset time interval and the interface performance prediction model, determine the corresponding message push queue according to the prediction result and a preset matching rule, and push the push data to the user side through the message push queue, thereby being able to improve the efficiency and flexibility of data push on the push platform.

[0056] To improve the efficiency and flexibility of data pushing on the pushing platform, this application provides an embodiment of a data pushing method. Refer to Figure 1 , the data pushing method specifically includes the following content:

[0057] Step S101: Collect the response time data of the user-side data pushing interface according to the time bucket aggregation technology, and collect the throughput rate data of the user-side data pushing interface according to the sliding window technology;

[0058] Optionally, in this embodiment, in order to solve the problem of low data pushing efficiency caused by inconsistent user interface performance, this step first collects the user-side interface performance data.

[0059] Optionally, the data affecting the interface performance is mainly RT / RPS (response time / throughput rate) data.

[0060] Optionally, in this embodiment, the time bucket aggregation technology divides time into fixed-size buckets (such as one bucket per minute), and aggregates and calculates the RT (response time) data within each bucket (such as calculating the average value, maximum value, minimum value, etc.). Through the time bucket aggregation technology, the system can more accurately capture the change trend of RT data and avoid the impact of instantaneous fluctuations on the system.

[0061] Optionally, in this embodiment, a sliding window (such as one window per 5 minutes, sliding once per minute) is used to perform real-time calculation on the RPS (throughput rate / number of requests per second) data to ensure that the system can dynamically track the changes in RPS. The sliding window technology can reflect the changes in RPS in real time, ensuring that the system can adjust the pushing strategy according to the latest RPS data.

[0062] Step S102: Decompose the response time data and the throughput rate data into a trend part and a residual part based on the time series decomposition technology, perform multi-dimensional feature extraction on the trend part and the residual part respectively, determine the corresponding trend features and residual features, combine the trend features and the residual features, perform feature selection on the feature vector after the feature combination according to the preset recursive feature elimination algorithm, determine the corresponding interface performance features, and train the initial model based on the interface performance features to determine the corresponding interface performance prediction model, where the trend features represent the long-term change trend features, and the residual features represent the random fluctuation features other than the trend features;

[0063] Optionally, in this embodiment, this step is a process of performing feature engineering on the RT / RPS data collected in step S101 and using the features to train the model.

[0064] Specifically, in traditional feature engineering, it is common to perform statistical feature extraction (such as mean, variance, maximum, minimum, etc.) on RT (response time) and RPS (requests per second / throughput). However, the limitation of this method is that it cannot capture complex patterns and trends in time series data.

[0065] Optionally, in this embodiment, in order to improve the prediction accuracy of the system and the accuracy of queue adjustment, multi-dimensional feature extraction based on time series decomposition is adopted.

[0066] First, based on the time series, the RT and RPS data are decomposed into two components, including the trend part and the residual part. Specifically, the Hodrick-Prescott filtering algorithm is used for time series decomposition, where:

[0067] Trend, representing the long-term change trend of the data.

[0068] Residual, representing the remaining part after removing the trend, including random fluctuations and noise.

[0069] Then, based on the time series decomposition, multi-dimensional features are extracted from the trend and the residual.

[0070] Trend features:

[0071] Trend slope: Calculate the slope of the trend part, reflecting the long-term change speed of RT and RPS.

[0072] Trend volatility: Calculate the standard deviation or variance of the trend part, reflecting the degree of fluctuation of the trend.

[0073] Trend extreme values: Extract the maximum and minimum values of the trend part, reflecting the extreme situations of RT and RPS.

[0074] Residual features:

[0075] Residual volatility: Calculate the standard deviation or variance of the residual part, reflecting the intensity of random fluctuations.

[0076] Residual autocorrelation: Calculate the autocorrelation coefficient of the residual part, reflecting the time dependence in the residual.

[0077] Residual extreme values: Extract the maximum and minimum values of the residual part, reflecting the extreme situations of random fluctuations.

[0078] Optionally, another time series-based decomposition method is to use the pywt library wavelet transform for data decomposition.

[0079] After extracting the above-mentioned trend features and residual features, the trend and residual features are combined to form a multi-dimensional feature vector. The Recursive Feature Elimination (RFE) algorithm is used to select the features that have the most influence on the prediction. In this way, deep-level information such as trends and residuals in RT and RPS data can be captured, covering the long-term trends, periodic changes, and random fluctuations of the data, enabling a more comprehensive description of the performance changes of the customer interface and significantly enhancing the expressive ability of the features.

[0080] Next, define a secondary fine-grained label, including a coarse-grained and a fine-fine-grained label.

[0081] The coarse-grained label is used to represent the RT and RPS values, and the fine-fine-grained label is used to represent the fluctuation levels of RT and RPS. The fine-fine-grained label is a more fine-grained performance metric based on the coarse-grained label.

[0082] For example, assume that we have collected the basic performance metrics of RT and RPS in the data collection module, including:

[0083] Response Time (RT): The average RT in the last 1 minute, the average RT in the last 5 minutes, the historical maximum, minimum, and variance of RT.

[0084] Requests Per Second (RPS): The average RPS in the last 1 minute, the average RPS in the last 5 minutes, the historical maximum, minimum, and variance of RPS.

[0085] Next, based on feature engineering, extract the corresponding features:

[0086] RT Trend: The change trend of RT in the past 5 minutes

[0087] RPS Trend: The change trend of RPS in the past 5 minutes

[0088] RT Residual: The difference between the actual value of RT and the trend prediction value.

[0089] RPS Residual: The difference between the actual value of RPS and the trend prediction value.

[0090] Construct the feature vector as [the average RT in the last 1 minute, the average RT in the last 5 minutes, the average RPS in the last 1 minute, the average RPS in the last 5 minutes, the slope of the RT trend, the slope of the RPS trend, the variance of the RT residual, the variance of the RPS residual]

[0091] Define the model secondary fine-grained label. The coarse-grained label is [the average RT value in the next 1 minute, the average RPS value in the next 1 minute]; the fine-fine-grained label is [the RT fluctuation level in the next 1 minute, the RPS fluctuation level in the next 1 minute]

[0092] Combine the above feature vectors with the secondary fine-grained labels to construct a dataset for training the initial model of the artificial intelligence model. The model can be selected from LSTM, XGBoost, random forest, etc. Use cross-validation to evaluate the model performance and adjust the hyperparameters to optimize the prediction accuracy. Through feature engineering and model training, the system can accurately predict the future performance of the interface and output the secondary fine-grained results of the interface performance, laying a foundation for subsequent communication queue adjustment based on the interface performance.

[0093] Step S103: Perform secondary fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model, determine the corresponding first interface performance prediction data and second interface performance prediction data, determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, determine the corresponding message push queue according to the first-level queue and the second-level queue, and push the push data to the user-side through the message push queue.

[0094] Optionally, in this embodiment, this step uses the interface performance prediction model obtained in step S102 to obtain the secondary prediction data of the user interface performance and perform dynamic adjustment of the message queue based on the rules.

[0095] Specifically, the system calls the interface performance prediction model at a preset time interval (such as every minute) to generate prediction data. The model outputs two key prediction values:

[0096] First interface performance prediction data (coarse and fine-grained interface performance data): Predict the RT and RPS values in the future period of time.

[0097] Second interface performance prediction data (fine and fine-grained interface performance data): Predict the fine-grained volatility of RT and RPS in the future period of time.

[0098] Perform queue adjustment based on the prediction values and the queue matching rules. Among them, the first-level queue is divided into four levels and divided according to the coarse and fine-grained interface performance data:

[0099] Level 1: RT < 50ms, RPS > 1000 / s (high-performance interface)

[0100] Level 2: 50ms ≤ RT < 100ms, 500 / s ≤ RPS < 1000 / s (medium-performance interface)

[0101] Level 3: 100ms ≤ RT < 200ms, 200 / s ≤ RPS < 500 / s (low-performance interface)

[0102] Level 3: RT ≥ 200 ms, RPS < 200 / s (extremely low-performance interface)

[0103] The secondary queue is divided into three levels, which are divided according to the performance data of the fine-grained interface:

[0104] Level X.1: RT variance < 10 ms, RPS fluctuation < 5% (high stability / low volatility).

[0105] Level X.2: 10 ms ≤ RT variance < 20 ms, 5% ≤ RPS fluctuation < 10% (medium stability / medium volatility).

[0106] Level X.3: RT variance ≥ 20 ms, RPS fluctuation ≥ 10% (low stability / high volatility)

[0107] The primary queue and the secondary queue form a nested structure. For example:

[0108] Level 1.1: High-performance interface, and low volatility of RT and RPS.

[0109] Level 1.2: High-performance interface, but medium volatility of RT and RPS.

[0110] Level 2.1: Medium-performance interface, and low volatility of RT and RPS.

[0111] Level 2.2: Medium-performance interface, but medium volatility of RT and RPS.

[0112] Illustrated by an example:

[0113] Suppose the system outputs the first interface performance prediction data: the RT value is 48 ms, and the RPS is 1100 / s. According to the primary queue matching rule, it belongs to Level 1.

[0114] Suppose the system outputs the second interface performance prediction data: the RT fluctuation is 5 ms, and the RPS fluctuation is 3%. According to the secondary queue matching rule, it belongs to Level X.1.

[0115] Combine the primary queue and the secondary queue to determine the final push queue Level 1.1.

[0116] Illustrated by another example:

[0117] Suppose the system outputs the first interface performance prediction data: the RT value is 78 ms, and the RPS is 300 / s. According to the primary queue matching rule, the RT value belongs to Level 2, while the RPS value belongs to Level 3. In this case, both take Level 3 as the final queue level.

[0118] It should be noted that the RT / RPS value range rule needs to be greater than 0, the value range is continuous, and the maximum marginal data needs to include infinity to ensure that the RT / RPS of the interface can fall into the corresponding queue.

[0119] Finally, after predicting the queue level, the system will transmit information according to the predicted queue level and monitor the real RT / RPS performance metrics at the client side during information transmission in real time.

[0120] Suppose the system prediction metrics are RT = 40ms, RPS = 1100 / s, with high RT and RPS stability, belonging to Level 1.1. The system transmits information according to Level 1.1. However, the real-time client-side performance metrics obtained after information transmission are RT = 60ms, RPS = 700 / s, with medium RT and RPS stability, belonging to Level 2.2. At this time, the information channel and the corresponding performance metric data are stored in the memory for subsequent model learning and real-time information channel adjustment.

[0121] The above is the technical implementation process for realizing this solution. The following introduces the system working process of this solution in a systematic way.

[0122] First, the data acquisition module collects the RT and RPS data of the customer interface in real time and stores them in the memory. The interface metrics correspond to each customer one by one.

[0123] Secondly, the data processing module interacts with the machine learning module to calculate and analyze the real-time RT and RPS data in the memory, predict the performance level data of the future user interface, and load the future performance level data into the memory.

[0124] Next, the queue management module dynamically adjusts the push queue level of the customer receipt according to the result of the data processing module and stores the configuration information in the memory.

[0125] Finally, the push module pushes the receipt data to the corresponding customer interface according to the instructions of the queue management module.

[0126] During the implementation process of the above system work, a monitoring and configuration module is introduced, which can interact with the above modules. The operation personnel can view the performance metrics of the customer interface in real time through the monitoring module and perform operations such as adding, deleting, modifying, and querying the configuration rules of the push queue through the configuration management module.

[0127] In addition, a new customer prediction module is additionally introduced. Based on historical data or data of similar customers, it predicts the initial performance of the new customer interface, and can immediately transfer the RT / RPS data of the new customer from the default queue to the appropriate queue after the data is collected. Through the new customer performance prediction and dynamic adjustment mechanism, the system can adapt to the access of new customers more quickly, increasing the efficiency and flexibility of data pushing.

[0128] This example shows how the data pushing queue is dynamically configured according to the user interface performance prediction in this embodiment.

[0129] As can be seen from the above description, the data pushing method provided by the embodiment of the present application can collect the response time data and throughput data of the user-side data pushing interface; based on the time series decomposition technology, decompose the response time data and throughput data into a trend part and a residual part, perform multi-dimensional feature extraction on the trend part and the residual part respectively, combine the trend features and the residual features, perform feature selection on the feature vector after feature combination according to the preset recursive feature elimination algorithm to determine the interface performance features, train the initial model based on the interface performance features to determine the corresponding interface performance prediction model; perform secondary fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model, determine the corresponding message pushing queue according to the prediction result and the preset matching rule, and push the push data to the user-side through the message pushing queue, thereby improving the efficiency and flexibility of data pushing on the push platform.

[0130] In an embodiment of the data pushing method of the present application, refer to Figure 2 and it may specifically include the following content:

[0131] Step S201: Divide time into multiple buckets of a fixed size;

[0132] Step S202: Perform aggregation calculation on the response time of the user-side data pushing interface within each of the buckets to determine the corresponding response time data.

[0133] Optionally, in this embodiment, this step is a time bucket aggregation technology, which divides time into buckets of a fixed size (such as one bucket per minute), and performs aggregation calculation (such as calculating the average value, maximum value, minimum value, etc.) on the RT (response time) data within each bucket. Through the time bucket aggregation technology, the system can more accurately capture the change trend of the RT data and avoid the impact of instantaneous fluctuations on the system.

[0134] Through step S202, this embodiment more accurately captures the change trend of the RT data and avoids the impact of instantaneous fluctuations on the system.

[0135] In an embodiment of the data pushing method of the present application, refer to Figure 3, it may also specifically include the following content:

[0136] Step S301: Based on the Hodrick-Prescott filtering technique, perform time-series data decomposition on the reaction time data to determine the corresponding reaction time trend and reaction time residual;

[0137] Step S302: Based on the Hodrick-Prescott filtering technique, perform time-series data decomposition on the throughput data to determine the corresponding throughput trend and throughput residual;

[0138] Step S303: Determine the corresponding trend part according to the reaction time trend and the throughput trend, and determine the corresponding residual part according to the reaction time residual and the throughput residual.

[0139] Optionally, in this embodiment, in order to improve the prediction accuracy of the system and the accuracy of queue adjustment, multi-dimensional feature extraction based on time-series decomposition is adopted.

[0140] First, decompose the RT and RPS data into two components based on time series, including the trend part and the residual part. Specifically, the Hodrick-Prescott filtering algorithm is used for time-series decomposition, where:

[0141] Trend, representing the long-term change trend of the data.

[0142] Residual, representing the remaining part after removing the trend, including random fluctuations and noise.

[0143] Then, based on the time-series decomposition, multi-dimensional features are extracted from the trend and the residual.

[0144] Trend features:

[0145] Trend slope: Calculate the slope of the trend part, reflecting the long-term change speed of RT and RPS.

[0146] Trend volatility: Calculate the standard deviation or variance of the trend part, reflecting the degree of fluctuation of the trend.

[0147] Trend extreme values: Extract the maximum and minimum values of the trend part, reflecting the extreme situations of RT and RPS.

[0148] Residual features:

[0149] Residual volatility: Calculate the standard deviation or variance of the residual part, reflecting the intensity of random fluctuations.

[0150] Residual autocorrelation: Calculate the autocorrelation coefficient of the residual part, reflecting the time dependence in the residual.

[0151] Residual extreme values: Extract the maximum and minimum values of the residual part, reflecting the extreme conditions of random fluctuations.

[0152] For example, assume that we have collected the basic performance metrics of RT and RPS in the data acquisition module.

[0153] Based on feature engineering, extract the corresponding trend features:

[0154] RT trend: The change trend of RT in the past 5 minutes

[0155] RPS trend: The change trend of RPS in the past 5 minutes

[0156] Based on feature engineering, extract the corresponding residual features:

[0157] RT residual: The difference between the actual value of RT and the trend prediction value.

[0158] RPS residual: The difference between the actual value of RPS and the trend prediction value.

[0159] Through step S303, in this embodiment, the trend features and residual features of the client interface performance extracted by feature engineering cover the long-term trend, periodic changes, and random fluctuations of the data, and can more comprehensively describe the performance changes of the client interface, significantly improving the expression ability of the features.

[0160] In an embodiment of the data push method of the present application, refer to Figure 4 , and it may specifically include the following content:

[0161] Step S401: Use polynomial fitting to perform multi-dimensional feature extraction on the trend part to determine the corresponding trend features, where the trend features include trend slope, trend volatility, and trend extreme values;

[0162] Step S402: Use the frequency domain analysis method to perform multi-dimensional feature extraction on the residual part to determine the corresponding residual features, where the residual features include residual volatility, residual autocorrelation, and residual extreme values.

[0163] Optionally, in this embodiment, based on time series decomposition, use polynomial fitting to extract multi-dimensional features from the trend part.

[0164] Specifically, polynomial fitting is to approximate the trend part of the time series by fitting a polynomial function.

[0165] The form of the polynomial function is: y = a0 + a1x + a2x 2 +...+ a n x n

[0166] Where x is the time variable, y is the time series value, a0, a1, …, an are polynomial coefficients, and n is the order of the polynomial.

[0167] Preferably, a second-order polynomial is selected, and the least squares method is used to fit the polynomial coefficients to minimize the error between the fitting curve and the time series data. The fitted polynomial function is used as the trend slope, trend volatility, and trend extreme value of the trend part, and is separated from the original time series.

[0168] Optionally, in this embodiment, based on the time series decomposition, a frequency domain analysis method is used to extract multi-dimensional features from the trend part.

[0169] Specifically, the numpy.fft is used for Fourier transform to extract multi-dimensional features from the residual part that can characterize its random fluctuation characteristics. This includes calculating the variance or standard deviation of the residual part to reflect the fluctuation degree of the residual; calculating the autocorrelation coefficient (ACF) of the residual part to reflect the correlation of the residual at different time lags; and extracting the maximum and minimum values of the residual part to reflect the abnormal fluctuation of the residual.

[0170] By performing time series decomposition, polynomial fitting, and frequency domain analysis to extract multi-dimensional features from the trend part and the residual part respectively, the long-term trend and short-term fluctuations of the time series can be comprehensively characterized, which has high flexibility and practicality.

[0171] Through step S402, this embodiment successfully extracts multi-dimensional features from the trend part and the residual part, laying a foundation for subsequent model training.

[0172] In an embodiment of the data push method of the present application, referring to Figure 5 , it may further specifically include the following content:

[0173] Step S501: Construct corresponding performance index labels based on the secondary fine-grained performance indicators;

[0174] Step S502: Input the interface performance characteristics and the performance index labels into the initial model for model training to determine the corresponding interface performance prediction model, which is used for secondary fine-grained performance prediction.

[0175] Optionally, in this embodiment, a secondary fine-grained label is defined, including a coarse-grained and a fine-grained level.

[0176] The coarse-grained label is used to represent the RT and RPS values, and the fine-grained label is used to represent the fluctuation levels of RT and RPS. The fine-grained level is a more detailed performance indicator based on the coarse-grained level.

[0177] For example, assume that we have collected the basic performance metrics of RT and RPS in the data collection module, including:

[0178] Response Time (RT): The average RT in the most recent 1 minute, the average RT in the most recent 5 minutes, the historical maximum, minimum, and variance of RT.

[0179] Requests Per Second (RPS): The average RPS in the most recent 1 minute, the average RPS in the most recent 5 minutes, the historical maximum, minimum, and variance of RPS.

[0180] Next, based on feature engineering, extract the corresponding features:

[0181] RT Trend: The change trend of RT in the past 5 minutes

[0182] RPS Trend: The change trend of RPS in the past 5 minutes

[0183] RT Residual: The difference between the actual value of RT and the trend prediction value.

[0184] RPS Residual: The difference between the actual value of RPS and the trend prediction value.

[0185] Construct the feature vector as [average RT in the most recent 1 minute, average RT in the most recent 5 minutes, average RPS in the most recent 1 minute, average RPS in the most recent 5 minutes, RT trend slope, RPS trend slope, RT residual variance, RPS residual variance]

[0186] Define the secondary fine-grained labels of the model. The coarse-grained labels are [average RT value within the next 1 minute, average RPS value within the next 1 minute]; the fine-grained labels are [RT fluctuation level within the next 1 minute, RPS fluctuation level within the next 1 minute]

[0187] Combine the above feature vector with the secondary fine-grained labels to construct a dataset for training the initial model of the artificial intelligence model. The model can be selected from LSTM, XGBoost, random forest, etc. Use cross-validation to evaluate the model performance and adjust the hyperparameters to optimize the prediction accuracy. Through feature engineering and model training, the system can accurately predict the future performance of the interface and output the secondary fine-grained results of the interface performance, laying a foundation for subsequent communication queue adjustment based on the interface performance.

[0188] Through step S502, this embodiment successfully determines the interface performance prediction model, laying a foundation for subsequent communication queue adjustment based on the interface performance.

[0189] In an embodiment of the data push method of this application, refer to Figure 6 , and it may specifically include the following content:

[0190] Step S601: Perform secondary fine-grained performance prediction on the client interface according to a preset time interval and the interface performance prediction model, and determine the corresponding first interface performance prediction data and second interface performance prediction data, where the first interface performance prediction data is used to represent the coarse-grained interface performance metrics, and the second interface performance prediction data is used to represent the fine-grained interface performance metrics;

[0191] Step S602: Determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, and determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, where the first-level queue and the second-level queue are in a nested structure.

[0192] Optionally, the system calls the interface performance prediction model at a preset time interval (such as every minute) to generate prediction data. The model outputs two key prediction values:

[0193] First interface performance prediction data (coarse-grained interface performance data): Predict the RT and RPS values in the future for a period of time.

[0194] Second interface performance prediction data (fine-grained interface performance data): Predict the fine-grained volatility of RT and RPS in the future for a period of time.

[0195] Based on the prediction values and the queue matching rules, perform queue adjustment, where the first-level queue is divided into four levels and divided according to the coarse-grained interface performance data:

[0196] Level 1: RT < 50ms, RPS > 1000 / s (high-performance interface)

[0197] Level 2: 50ms ≤ RT < 100ms, 500 / s ≤ RPS < 1000 / s (medium-performance interface)

[0198] Level 3: 100ms ≤ RT < 200ms, 200 / s ≤ RPS < 500 / s (low-performance interface)

[0199] Level 3: RT ≥ 200ms, RPS < 200 / s (extremely low-performance interface)

[0200] The second-level queue is divided into three levels and divided according to the fine-grained interface performance data:

[0201] Level X.1: RT variance < 10ms, RPS fluctuation < 5% (high stability / low volatility).

[0202] Level X.2: 10ms ≤ RT variance < 20ms, 5% ≤ RPS fluctuation < 10% (medium stability / medium volatility).

[0203] Level X.3: The RT variance ≥ 20 ms, and the RPS fluctuation ≥ 10% (low stability / high volatility)

[0204] The first-level queue and the second-level queue form a nested structure. For example:

[0205] Level 1.1: High-performance interface, and low volatility of RT and RPS.

[0206] Level 1.2: High-performance interface, but medium volatility of RT and RPS.

[0207] Level 2.1: Medium-performance interface, and low volatility of RT and RPS.

[0208] Level 2.2: Medium-performance interface, but medium volatility of RT and RPS.

[0209] Through step S602, in this embodiment, dynamic queue division is successfully performed based on the performance metrics predicted by the model.

[0210] In an embodiment of the data push method of the present application, refer to Figure 7 , and it may specifically include the following content:

[0211] Step S701: Determine the overall priority of the corresponding message push according to the first-level queue, and determine the fine-grained priority of the corresponding message push according to the second-level queue;

[0212] Step S702: Determine the corresponding message push queue according to the overall priority of the message push and the fine-grained priority of the message push.

[0213] Optionally, in this embodiment, the first-level queue and the second-level queue form a nested structure.

[0214] When performing queue priority division, first determine the overall priority based on the first-level queue, and then determine the fine-grained priority within the first-level priority based on the second-level queue.

[0215] Illustrated with an example of an embodiment:

[0216] Suppose the system outputs the first interface performance prediction data: the RT value is 48 ms, and the RPS is 1100 / s. According to the first-level queue matching rule, it belongs to Level 1.

[0217] Suppose the system outputs the second interface performance prediction data: the RT volatility is 5 ms, and the RPS volatility is 3%. According to the second-level queue matching rule, it belongs to Level X.1.

[0218] Combine the first-level queue and the second-level queue to determine the final push queue Level 1.1.

[0219] Illustrate with another embodiment:

[0220] Suppose the system outputs the first interface performance prediction data: the RT value is 78 ms, and the RPS is 300 / s. According to the first-level queue matching rule, the RT value belongs to Level 2, and the RPS value belongs to Level 3. In this case, Level 3 is taken as the final queue level for both.

[0221] Through step S702, this embodiment successfully determines the message push queue according to the second-level priority division mechanism, improving the efficiency and flexibility of message push.

[0222] To improve the efficiency and flexibility of data push on the push platform, this application provides an embodiment of a data push device for implementing all or part of the content of the data push method. Refer to Figure 8 The data push device specifically includes the following:

[0223] The data acquisition module 10 is used to collect the response time data of the user-side data push interface according to the time bucket aggregation technology, and collect the throughput data of the user-side data push interface according to the sliding window technology;

[0224] The machine learning module 20 is used to decompose the response time data and the throughput data into a trend part and a residual part based on the time series decomposition technology, perform multi-dimensional feature extraction on the trend part and the residual part respectively, determine the corresponding trend features and residual features, combine the trend features and the residual features, perform feature selection on the feature vector after the feature combination according to the preset recursive feature elimination algorithm, determine the corresponding interface performance features, and train the initial model based on the interface performance features to determine the corresponding interface performance prediction model, where the trend features represent the long-term change trend features, and the residual features represent the random fluctuation features outside the trend features;

[0225] The queue management module 30 is used to perform second-level fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model, determine the corresponding first interface performance prediction data and second interface performance prediction data, determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, determine the corresponding message push queue according to the first-level queue and the second-level queue, and push the push data to the user-side through the message push queue.

[0226] As can be seen from the above description, the data push device provided by the embodiments of the present application can collect the response time data and throughput data of the user-side data push interface; decompose the response time data and throughput data into a trend part and a residual part based on the time series decomposition technology, perform multi-dimensional feature extraction on the trend part and the residual part respectively, combine the trend features and the residual features, perform feature selection on the feature vector after feature combination according to the preset recursive feature elimination algorithm, determine the interface performance features, train an initial model based on the interface performance features, and determine the corresponding interface performance prediction model; perform secondary fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model, determine the corresponding message push queue according to the prediction result and the preset matching rule, and push the push data to the user side through the message push queue, thereby improving the efficiency and flexibility of data push on the push platform.

[0227] From a hardware perspective, in order to improve the efficiency and flexibility of data push on the push platform, the present application provides an embodiment of an electronic device for implementing all or part of the content in the data push method. The electronic device specifically includes the following:

[0228] A processor, a memory, a communications interface, and a bus; wherein, the processor, the memory, and the communications interface complete mutual communication through the bus; the communications interface is used to implement information transmission between the data push method and related devices such as the core business system, the user terminal, and the related database. The logic controller can be a desktop computer, a tablet computer, a mobile terminal, etc., and this embodiment is not limited thereto. In this embodiment, the logic controller can be implemented with reference to the embodiments of the data push method in the embodiments, and the embodiments of the data push method are incorporated herein, and the repeated parts will not be described again.

[0229] It can be understood that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0230] In practical applications, part of the data push method can be executed on the electronic device side as described above, or all operations can be completed in the client device. Specifically, it can be selected according to the processing capacity of the client device and the limitations of the user usage scenario. The present application does not limit this. If all operations are completed in the client device, the client device may further include a processor.

[0231] The above-mentioned client device may have a communication module (i.e., communication unit), which can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side. In other implementation scenarios, it may also include a server of an intermediate platform, such as a server of a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, or a server cluster composed of multiple servers, or a server structure of a distributed device.

[0232] Figure 9 It is a schematic block diagram of the system composition of the electronic device 9600 according to an embodiment of the present application. As Figure 9 shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It should be noted that this Figure 9 is exemplary; other types of structures can also be used to supplement or replace this structure to achieve telecommunication functions or other functions.

[0233] In one embodiment, the data push method function can be integrated into the central processing unit 9100. Among them, the central processing unit 9100 can be configured to perform the following controls:

[0234] Step S101: Collect the response time data of the user-side data push interface according to the time bucket aggregation technology, and collect the throughput data of the user-side data push interface according to the sliding window technology;

[0235] Step S102: Based on the time series decomposition technology, decompose the response time data and the throughput data into a trend part and a residual part, perform multi-dimensional feature extraction on the trend part and the residual part respectively, determine the corresponding trend features and residual features, combine the trend features and the residual features, perform feature selection on the feature vector after the feature combination according to the preset recursive feature elimination algorithm, determine the corresponding interface performance features, and train an initial model based on the interface performance features to determine the corresponding interface performance prediction model, where the trend features represent the long-term change trend features, and the residual features represent the random fluctuation features other than the trend features;

[0236] Step S103: Perform secondary fine-grained performance prediction on the user-side interface according to a preset time interval and the interface performance prediction model, determine the corresponding first interface performance prediction data and second interface performance prediction data, determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, determine the corresponding message push queue according to the first-level queue and the second-level queue, and push the push data to the user-side through the message push queue.

[0237] As can be seen from the above description, the electronic device provided by the embodiment of the present application collects the response time data and throughput data of the user-side data push interface; based on the time series decomposition technology, decomposes the response time data and throughput data into a trend part and a residual part, performs multi-dimensional feature extraction on the trend part and the residual part respectively, combines the trend features and the residual features, performs feature selection on the feature vector after feature combination according to the preset recursive feature elimination algorithm, determines the interface performance features, trains an initial model based on the interface performance features, and determines the corresponding interface performance prediction model; performs secondary fine-grained performance prediction on the user-side interface according to a preset time interval and the interface performance prediction model, determines the corresponding message push queue according to the prediction result and the preset matching rule, and pushes the push data to the user-side through the message push queue, thereby improving the efficiency and flexibility of data push on the push platform.

[0238] In another embodiment, the data push method can be separately configured from the central processing unit 9100. For example, the data push method can be configured as a chip connected to the central processing unit 9100, and the function of the data push method is realized through the control of the central processing unit.

[0239] As Figure 9 shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It should be noted that the electronic device 9600 does not necessarily have to include Figure 9 all the components shown in Figure 9 ; in addition, the electronic device 9600 may further include

[0240] As Figure 9 shown, the central processing unit 9100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor devices and / or logic devices. The central processing unit 9100 receives inputs and controls the operations of the various components of the electronic device 9600.

[0241] Among them, the memory 9140 can be, for example, one or more of a buffer, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory, or other suitable devices. The above-mentioned information related to failures can be stored, and in addition, a program for executing relevant information can also be stored. And the central processing unit 9100 can execute the program stored in the memory 9140 to implement information storage or processing, etc.

[0242] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to supply power to the electronic device 9600. The display 9160 is used to display display objects such as images and texts. The display can be, for example, an LCD display, but is not limited thereto.

[0243] The memory 9140 can be a solid-state memory. For example, a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It can also be a memory that stores information even when powered off, can be selectively erased and has more data. An example of this memory is sometimes called an EPROM, etc. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes called a buffer). The memory 9140 can include an application / function storage unit 9142, which is used to store application programs and function programs or the processes for operating the electronic device 9600 through the central processing unit 9100.

[0244] The memory 9140 can also include a data storage unit 9143, which is used to store data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 can include various drivers of the electronic device for communication functions and / or for performing other functions of the electronic device (such as a messaging application, an address book application, etc.).

[0245] The communication module 9110 is a transmitter / receiver that transmits and receives signals via the antenna 9111. The communication module 9110 is coupled to the central processing unit 9100 to provide input signals and receive output signals, which can be the same as in the case of a conventional mobile communication terminal.

[0246] Based on different communication technologies, in the same electronic device, multiple communication modules 9110 can be provided, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module, etc. The communication module 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide an audio output via the speaker 9131 and receive an audio input from the microphone 9132, thereby implementing normal telecommunication functions. The audio processor 9130 can include any suitable buffers, decoders, amplifiers, etc. Additionally, the audio processor 9130 is also coupled to a central processor 9100, enabling recording on the device through the microphone 9132 and playing back the sound stored on the device through the speaker 9131.

[0247] Embodiments of the present application also provide a computer-readable storage medium capable of implementing all steps in the data push method where the execution subject in the above embodiments is a server or a client. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, all steps in the data push method where the execution subject in the above embodiments is a server or a client are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0248] Step S101: Collect response time data of the user-side data push interface according to the time binning aggregation technology, and collect throughput data of the user-side data push interface according to the sliding window technology;

[0249] Step S102: Decompose the response time data and the throughput data into a trend part and a residual part based on the time series decomposition technology. Perform multi-dimensional feature extraction on the trend part and the residual part respectively to determine the corresponding trend features and residual features. Combine the trend features and the residual features. Perform feature selection on the feature vector after the feature combination according to a preset recursive feature elimination algorithm to determine the corresponding interface performance features. Train an initial model based on the interface performance features to determine the corresponding interface performance prediction model, where the trend features represent long-term change trend features, and the residual features represent random fluctuation features other than the trend features;

[0250] Step S103: Perform secondary fine-grained performance prediction on the user-side interface according to a preset time interval and the interface performance prediction model to determine the corresponding first interface performance prediction data and second interface performance prediction data. Determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule. Determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule. Determine the corresponding message push queue according to the first-level queue and the second-level queue. Push the push data to the user-side through the message push queue.

[0251] As can be seen from the above description, the computer-readable storage medium provided by the embodiments of the present application collects the response time data and throughput data of the user-side data push interface; decomposes the response time data and throughput data into a trend part and a residual part based on the time series decomposition technology, performs multi-dimensional feature extraction on the trend part and the residual part respectively, combines the trend features and the residual features, performs feature selection on the feature vector after the feature combination according to the preset recursive feature elimination algorithm, determines the interface performance features, trains an initial model based on the interface performance features, and determines the corresponding interface performance prediction model; performs secondary fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model, determines the corresponding message push queue according to the prediction result and the preset matching rule, and pushes the push data to the user side through the message push queue, thereby improving the efficiency and flexibility of data push on the push platform.

[0252] An embodiment of the present application also provides a computer program product capable of implementing all steps in the data push method in which the execution subject in the above embodiments is a server or a client. When the computer program / instructions are executed by a processor, the steps of the data push method are implemented. For example, the computer program / instructions implement the following steps:

[0253] Step S101: Collect the response time data of the user-side data push interface according to the time binning aggregation technology, and collect the throughput data of the user-side data push interface according to the sliding window technology;

[0254] Step S102: Decompose the response time data and the throughput data into a trend part and a residual part based on the time series decomposition technology, perform multi-dimensional feature extraction on the trend part and the residual part respectively, determine the corresponding trend features and residual features, combine the trend features and the residual features, perform feature selection on the feature vector after the feature combination according to the preset recursive feature elimination algorithm, determine the corresponding interface performance features, train an initial model based on the interface performance features, and determine the corresponding interface performance prediction model, where the trend features represent long-term change trend features, and the residual features represent random fluctuation features other than the trend features;

[0255] Step S103: Perform secondary fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model, determine the corresponding first interface performance prediction data and second interface performance prediction data, determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, determine the corresponding message push queue according to the first-level queue and the second-level queue, and push the push data to the user-side through the message push queue.

[0256] As can be seen from the above description, the computer program product provided by the embodiment of the present application collects the response time data and throughput data of the user-side data push interface; based on the time series decomposition technology, decomposes the response time data and throughput data into a trend part and a residual part, respectively performs multi-dimensional feature extraction on the trend part and the residual part, combines the trend features and the residual features, performs feature selection on the feature vector after feature combination according to the preset recursive feature elimination algorithm, determines the interface performance features, trains an initial model based on the interface performance features, and determines the corresponding interface performance prediction model; performs secondary fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model, determines the corresponding message push queue according to the prediction result and the preset matching rule, and pushes the push data to the user-side through the message push queue, thereby improving the efficiency and flexibility of data push on the push platform.

[0257] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a device, or a computer program product. Therefore, the present invention can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0258] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0259] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements in the process Figure 1 one process or multiple processes and / or blocks Figure 1 the functions specified in one block or multiple blocks.

[0260] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, such that a series of operational steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0261] Specific embodiments are used in the present invention to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, based on the idea of the present invention, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A data push method, characterized in that The method includes: Collecting the response time data of the user - side data push interface according to the time - bucket aggregation technology, and collecting the throughput data of the user - side data push interface according to the sliding window technology; Based on the time - series decomposition technology, decomposing the response time data and the throughput data into a trend part and a residual part, respectively performing multi - dimensional feature extraction on the trend part and the residual part, determining the corresponding trend features and residual features, combining the trend features and the residual features, performing feature selection on the feature vector after the feature combination according to the preset recursive feature elimination algorithm, determining the corresponding interface performance features, and training an initial model based on the interface performance features to determine the corresponding interface performance prediction model, wherein the trend features represent the long - term change trend features, and the residual features represent the random fluctuation features other than the trend features; Performing secondary fine - grained performance prediction on the user - side interface according to the preset time interval and the interface performance prediction model, determining the corresponding first interface performance prediction data and second interface performance prediction data, determining the corresponding first - level queue according to the first interface performance prediction data and the first message queue matching rule, determining the corresponding second - level queue according to the second interface performance prediction data and the second message queue matching rule, determining the corresponding message push queue according to the first - level queue and the second - level queue, and pushing the push data to the user - side through the message push queue.

2. The data push method according to claim 1, wherein The collecting the response time data of the user - side data push interface according to the time - bucket aggregation technology includes: Dividing time into multiple buckets of a fixed size; Performing aggregation calculation on the response time of the user - side data push interface within each bucket to determine the corresponding response time data.

3. The data push method according to claim 1, wherein The decomposing the response time data and the throughput data into a trend part and a residual part based on the time - series decomposition technology includes: Performing time - series data decomposition on the response time data based on the Hodrick - Prescott filtering technology to determine the corresponding response time trend and response time residual; Performing time - series data decomposition on the throughput data based on the Hodrick - Prescott filtering technology to determine the corresponding throughput trend and throughput residual; Determining the corresponding trend part according to the response time trend and the throughput trend, and determining the corresponding residual part according to the response time residual and the throughput residual.

4. The data push method according to claim 1, wherein The respectively performing multi - dimensional feature extraction on the trend part and the residual part to determine the corresponding trend features and residual features includes: Performing multi - dimensional feature extraction on the trend part using polynomial fitting to determine the corresponding trend features, wherein the trend features include trend slope, trend volatility, and trend extreme value; Performing multi - dimensional feature extraction on the residual part using the frequency - domain analysis method to determine the corresponding residual features, wherein the residual features include residual volatility, residual autocorrelation, and residual extreme value.

5. The data push method according to claim 1, wherein The training an initial model based on the interface performance features to determine the corresponding interface performance prediction model includes: Construct corresponding performance metric labels based on secondary fine-grained performance metrics; Input the interface performance characteristics and the performance metric labels into an initial model for model training to determine a corresponding interface performance prediction model, which is used for secondary fine-grained performance prediction.

6. The data push method according to claim 1, characterized in that Performing secondary fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model to determine corresponding first interface performance prediction data and second interface performance prediction data, and determining a corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, and determining a corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, includes: Performing secondary fine-grained performance prediction on the user-side interface according to the preset time interval and the interface performance prediction model to determine corresponding first interface performance prediction data and second interface performance prediction data, where the first interface performance prediction data is used to represent coarse-fine-grained interface performance metrics, and the second interface performance prediction data is used to represent fine-fine-grained interface performance metrics; Determining a corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, and determining a corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, where the first-level queue and the second-level queue are nested structures.

7. The data pushing method according to claim 1, characterized in that Determining a corresponding message push queue according to the first-level queue and the second-level queue, includes: Determining a corresponding overall message push priority according to the first-level queue, and determining a corresponding fine-grained message push priority according to the second-level queue; Determining a corresponding message push queue according to the overall message push priority and the fine-grained message push priority.

8. A data push device, characterized in that, The device includes: A data acquisition module, configured to collect response time data of the user-side data push interface according to the time bucket aggregation technology, and collect throughput data of the user-side data push interface according to the sliding window technology; A machine learning module, configured to decompose the response time data and the throughput data into a trend part and a residual part based on the time series decomposition technology, perform multi-dimensional feature extraction on the trend part and the residual part respectively to determine corresponding trend features and residual features, combine the trend features and the residual features, perform feature selection on the feature vector after the feature combination according to the preset recursive feature elimination algorithm to determine corresponding interface performance characteristics, and train an initial model based on the interface performance characteristics to determine a corresponding interface performance prediction model, where the trend feature represents the long-term change trend feature, and the residual feature represents the random fluctuation feature outside the trend feature; The queue management module is used to perform secondary fine-grained performance prediction on the client interface according to a preset time interval and the interface performance prediction model, determine the corresponding first interface performance prediction data and second interface performance prediction data, determine the corresponding first-level queue according to the first interface performance prediction data and the first message queue matching rule, determine the corresponding second-level queue according to the second interface performance prediction data and the second message queue matching rule, determine the corresponding message push queue according to the first-level queue and the second-level queue, and push the push data to the client through the message push queue.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the data push method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the data push method according to any one of claims 1 to 7.

Images