A data processing method, apparatus, device, and readable storage medium

Abstract

This application discloses a data processing method, apparatus, device, and readable storage medium, relating to the field of communication technology, to improve the accuracy of obtained QoE levels. The method includes: acquiring structured data; generating supplementary sample data based on an improved deep convolutional generative adversarial network (DCGAN) and the structured data; wherein the supplementary sample data is generated based on the mapping results of target information from the structured data to different spatial representations of an image, and the target information includes clustering results of structured indices of the structured data. Embodiments of this application can improve the accuracy of obtained QoE levels.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a data processing method, apparatus, device, and readable storage medium. Background Technology

[0002] With the arrival of the 5G era, video transmission accounts for a large portion of data traffic, making it crucial for ensuring a good user experience for video services. However, operators cannot directly perceive the user's QoE (Quality of Experience) from the client side; they can only monitor the user's subjective experience through network-side KPIs (Key Performance Indicators). Traditional methods for assessing QoE levels involve weighted summation of several KPIs, but this approach yields inaccurate QoE ratings.

[0003] To improve the accuracy of obtained QoE ratings, existing technologies have proposed AI (Artificial Intelligence)-based video experience quality monitoring methods. Currently, the proportion of poor-quality data is relatively small, and data from existing operator networks is difficult to obtain, posing a significant challenge to AI-based quality monitoring methods.

[0004] Currently, the main ways to address imbalanced samples include the following two:

[0005] One approach is random replication, which involves replicating a small number of low-quality samples to achieve sample balance. Another approach is sample construction methods, such as using SMOTE (Synthetic minority over-sampling technique) and its derivative algorithms.

[0006] However, the above methods have the problem of inaccurate QoE level assessment. Summary of the Invention

[0007] This application provides a data processing method, apparatus, device, and readable storage medium to improve the accuracy of the obtained QoE rating.

[0008] In a first aspect, embodiments of this application provide a data processing method, including:

[0009] Obtain structured data;

[0010] Supplementary sample data is generated based on the improved DCGAN (Deep Convolutional Generative Adversarial Networks) and the structured data.

[0011] The supplementary sample data is generated based on the mapping results of the target information of the structured data to different spaces of the image, and the target information includes the clustering results of the structured indicators of the structured data.

[0012] The acquisition of structured data includes:

[0013] Acquire data to be processed, wherein the data to be processed includes: first data from a video terminal testing tool and second data from a network-side data source, wherein the first data includes video experience measurement indicators and a first related indicator; and the second data includes video-related indicators and a second related indicator.

[0014] Matching data is obtained from the data to be processed, and the matching data is used as the structured data.

[0015] Related metrics include: user IP, video ID or URL (Uniform Resource Locator), video playback start time, and video playback end time.

[0016] Obtaining matching data from the data to be processed includes:

[0017] Preprocessing the first data and the second data yields the target first data and the target second data;

[0018] The target first data and the target second data are matched using a quadruple to obtain matching data;

[0019] The quadruple includes the user IP, video ID, video playback start time difference, and video playback end time difference.

[0020] The step of matching the first target data and the second target data using quadruples to obtain matching data includes:

[0021] Based on the user IP and video ID in the first and second association indicators, the target first data and the target second data are matched to obtain first intermediate data from the target first data and second intermediate data from the target second data, wherein the first intermediate data and the second intermediate data correspond to the same user IP and video ID;

[0022] Based on the video playback start time and video playback end time in the first correlation index corresponding to the first intermediate data and the video playback start time and video playback end time in the second correlation index corresponding to the second intermediate data, the first intermediate data and the second intermediate data are matched to obtain third intermediate data from the target first data and fourth intermediate data from the target second data. The difference between the video playback start time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback start time difference, and the difference between the video playback end time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback end time difference.

[0023] The third intermediate data and the fourth intermediate data are used as the matching data.

[0024] Specifically, the mapping of the target information of the structured data to different spaces of the image is performed in the following manner:

[0025] The structured indicators of the structured data are clustered to obtain L types of structured indicators;

[0026] Based on the correspondence between structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L-type structured indicators is obtained.

[0027] Based on the pixel range corresponding to each type of structured indicator, map each type of structured indicator to the image;

[0028] Where N is an integer greater than or equal to 1.

[0029] The step of generating supplementary sample data based on the improved deep convolutional generative adversarial network (DCGAN) and the structured data includes:

[0030] The fourth data is obtained through the generator of the DCGAN;

[0031] The fourth data and the target structured data are processed by the discriminator of the DCGAN to obtain the fifth data, wherein the target structured data is the structured data corresponding to the target label; the proportion of the target structured data in the total amount of structured data is less than the proportion of other structured data under the target label in the total amount of structured data; in the discriminator, the structured indicators of the fourth data and the target structured data are mapped to different spaces of the image;

[0032] Calculate the loss function for the fourth and fifth data;

[0033] The generator and the discriminator are repeatedly trained iteratively according to the loss function to obtain a data balancing model, and the output of the trained data balancing model is used as the supplementary sample data.

[0034] The process of obtaining the fourth data through the generator of the DCGAN includes:

[0035] Randomly generate noise data 1×H, where H represents the dimension of the noise data, and the dimension of the noise data is smaller than the dimension of the structured data;

[0036] The noise data is processed by a generator to obtain a fourth data point with 1×N dimensions.

[0037] The step of passing the noise data through a generator to obtain 1×N dimensional fourth data includes:

[0038] The noise data is input into the fully connected layer of the generator, into a 1×N dimensional third data, where N represents the number of indicators in the structured data, which is an integer greater than 0;

[0039] The 1×N dimensional third data is input into the deconvolution layer of the generator to obtain M×M dimensional data;

[0040] The M×M dimensional data is input into the reshaping layer and clipping layer of the generator to obtain the 1×N dimensional fourth data, where M is... Round up.

[0041] The step of passing the fourth data and the target structured data through the discriminator of the DCGAN to obtain the fifth data includes:

[0042] The 1×N dimensional fourth data and the target structured data are processed by a discriminator to obtain the 1×N dimensional fifth data.

[0043] The step of passing the 1×N dimensional fourth data and the target structured data through a discriminator to obtain the 1×N dimensional fifth data includes:

[0044] The 1×N dimensional fourth data and the target structured data are input into the discriminator and mapped into an M×M dimensional target image;

[0045] The spatial characteristics of the target image are extracted using the convolutional layer of the discriminator;

[0046] The output of the convolutional layer of the discriminator is passed through the fully connected layer of the discriminator to obtain the 1×N dimensional fifth data.

[0047] The step of inputting the 1×N dimensional fourth data and the target structured data into the discriminator and mapping them into an M×M dimensional target image includes:

[0048] Obtain indicator classification information;

[0049] Based on the index classification information, the N indicators of the 1×N dimensional fourth data are mapped to different spaces of the image to obtain an M×M dimensional target image.

[0050] The index classification information is obtained by clustering the structured indicators of the structured data.

[0051] The acquisition of indicator classification information includes:

[0052] The structured indicators corresponding to the structured data are clustered to obtain n clustering indicators, and the information of the n clustering indicators is used as the indicator classification information.

[0053] Where n is an integer and n≥2.

[0054] The method further includes, after generating the supplementary sample data:

[0055] The video quality monitoring model is trained using the target structured data and the supplementary sample data.

[0056] The method further includes, after training the video quality monitoring model, the following:

[0057] The Quality of Experience (QoE) level is obtained by using third-party data obtained from network-side data sources and a trained video quality monitoring model.

[0058] Secondly, embodiments of this application also provide a data processing apparatus, including:

[0059] The first acquisition module is used to acquire structured data;

[0060] The first generation module is used to generate supplementary sample data based on the improved DCGAN and the structured data;

[0061] The supplementary sample data is generated based on the mapping results of the target information of the structured data to different spaces of the image, and the target information includes the clustering results of the structured indicators of the structured data.

[0062] The first acquisition module includes:

[0063] The first acquisition submodule is used to acquire data to be processed, wherein the data to be processed includes: first data from the video terminal testing tool and second data from the network-side data source. The first data includes video experience measurement indicators and first related indicators; the second data includes video-related indicators and second related indicators.

[0064] The second acquisition submodule is used to obtain matching data from the data to be processed and use the matching data as the structured data.

[0065] The relevant metrics include: user IP, video ID or URL, video playback start time, and video playback end time.

[0066] The second acquisition submodule includes:

[0067] The preprocessing unit is used to preprocess the first data and the second data to obtain target first data and target second data;

[0068] A matching unit is used to match the first target data and the second target data using a quadruple to obtain matching data;

[0069] The quadruple includes the user IP, video ID, video playback start time difference, and video playback end time difference.

[0070] The matching unit includes:

[0071] The first matching subunit is configured to match the target first data and the target second data according to the user IP and video ID in the first association index and the second association index, obtain first intermediate data from the target first data and second intermediate data from the target second data, wherein the first intermediate data and the second intermediate data correspond to the same user IP and video ID;

[0072] The second matching subunit is used to match the first intermediate data and the second intermediate data according to the video playback start time and video playback end time in the first correlation index corresponding to the first intermediate data and the video playback start time and video playback end time in the second correlation index corresponding to the second intermediate data, to obtain the third intermediate data from the target first data and the fourth intermediate data from the target second data, wherein the difference between the video playback start time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback start time difference, and the difference between the video playback end time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback end time difference;

[0073] The first acquisition subunit is used to use the third intermediate data and the fourth intermediate data as the matching data.

[0074] Specifically, the mapping of the target information of the structured data to different spaces of the image is performed in the following manner:

[0075] The structured indicators of the structured data are clustered to obtain L types of structured indicators;

[0076] Based on the correspondence between structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L-type structured indicators is obtained.

[0077] Based on the pixel range corresponding to each type of structured indicator, map each type of structured indicator to the image;

[0078] Where L is an integer greater than or equal to 1.

[0079] The first generation module includes:

[0080] The first acquisition submodule is used to obtain the fourth data through the generator of the DCGAN;

[0081] The second acquisition submodule is used to process the fourth data and the target structured data through the discriminator of the DCGAN to obtain the fifth data, wherein the target structured data is the structured data corresponding to the target label; the proportion of the target structured data in the total amount of structured data is less than the proportion of other structured data under the target label in the total amount of structured data; in the discriminator, the structured indicators of the fourth data and the target structured data are mapped to different spaces of the image;

[0082] The first calculation submodule is used to calculate the loss function of the fourth data and the fifth data;

[0083] The third acquisition submodule is used to repeatedly iterate and train the generator and the discriminator according to the loss function to obtain the data balance processing model, and use the output of the trained data balance processing model as the supplementary sample data.

[0084] The first acquisition submodule includes:

[0085] The first generation unit is used to randomly generate noise data 1×H, where H represents the dimension of the noise data, and the dimension of the noise data is smaller than the dimension of the structured data.

[0086] The second generation unit is used to process the noise data through a generator to obtain 1×N dimensional fourth data.

[0087] The second generation unit includes:

[0088] The first generation subunit is used to input the noise data into the fully connected layer of the generator, to the 1×N dimensional third data, where N represents the number of indicators of the structured data, which is an integer greater than 0;

[0089] The second generation subunit is used to input the 1×N dimensional third data into the deconvolution layer of the generator to obtain M×M dimensional data;

[0090] The third generation subunit is used to input the M×M dimensional data into the reshaping layer and clipping layer of the generator to obtain the 1×N dimensional fourth data, where M is... Round up.

[0091] The second acquisition submodule is used to pass the 1×N-dimensional fourth data and the target structured data through a discriminator to obtain 1×N-dimensional fifth data.

[0092] The second acquisition submodule includes:

[0093] The first generation unit is used to input the 1×N dimensional fourth data and the target structured data into the discriminator and map them into an M×M dimensional target image.

[0094] The first extraction unit is used to extract the spatial characteristics of the target image using the convolutional layer of the discriminator;

[0095] The second generation unit is used to obtain the 1×N-dimensional fifth data by passing the output of the convolutional layer of the discriminator through the fully connected layer of the discriminator.

[0096] The first generation unit includes:

[0097] The first acquisition subunit is used to acquire indicator classification information;

[0098] The first mapping subunit is used to map the N indicators of the 1×N-dimensional fourth data to different spaces of the image according to the indicator classification information, so as to obtain an M×M-dimensional target image.

[0099] The index classification information is obtained by clustering the structured indicators of the structured data.

[0100] The first acquisition subunit is used to cluster the structured indicators corresponding to the structured data to obtain n clustering indicators, and use the information of the n clustering indicators as the indicator classification information.

[0101] Where n is an integer and n≥2.

[0102] The device further includes:

[0103] The training module is used to train the video quality monitoring model using the target structured data and the supplementary sample data.

[0104] The device further includes:

[0105] The second acquisition module is used to obtain the QoE level by using third data obtained from the network-side data source and the trained video quality detection model.

[0106] Thirdly, embodiments of this application provide a data processing apparatus, including: a processor and a transceiver; wherein the processor is used for:

[0107] Obtain structured data;

[0108] Supplementary sample data is generated based on the improved DCGAN and the structured data;

[0109] The supplementary sample data is generated based on the mapping results of the target information of the structured data to different spaces of the image, and the target information includes the clustering results of the structured indicators of the structured data.

[0110] The processor is further configured to:

[0111] Acquire data to be processed, wherein the data to be processed includes: first data from a video terminal testing tool and second data from a network-side data source, wherein the first data includes video experience measurement indicators and a first related indicator; and the second data includes video-related indicators and a second related indicator.

[0112] Matching data is obtained from the data to be processed, and the matching data is used as the structured data.

[0113] The associated metrics include: user IP, video ID or URL, video playback start time, and video playback end time; the processor is also used for:

[0114] Preprocessing the first data and the second data yields the target first data and the target second data;

[0115] The target first data and the target second data are matched using a quadruple to obtain matching data;

[0116] The quadruple includes the user IP, video ID, video playback start time difference, and video playback end time difference.

[0117] The processor is further configured to:

[0118] Based on the user IP and video ID in the first and second association indicators, the target first data and the target second data are matched to obtain first intermediate data from the target first data and second intermediate data from the target second data, wherein the first intermediate data and the second intermediate data correspond to the same user IP and video ID;

[0119] Based on the video playback start time and video playback end time in the first correlation index corresponding to the first intermediate data and the video playback start time and video playback end time in the second correlation index corresponding to the second intermediate data, the first intermediate data and the second intermediate data are matched to obtain third intermediate data from the target first data and fourth intermediate data from the target second data. The difference between the video playback start time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback start time difference, and the difference between the video playback end time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback end time difference.

[0120] The third intermediate data and the fourth intermediate data are used as the matching data.

[0121] The processor is further configured to: map the target information of the structured data to different spaces of the image in the following manner:

[0122] The structured indicators of the structured data are clustered to obtain L types of structured indicators;

[0123] Based on the correspondence between structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L-type structured indicators is obtained.

[0124] Based on the pixel range corresponding to each type of structured indicator, map each type of structured indicator to the image;

[0125] Where L is an integer greater than or equal to 1.

[0126] The processor is further configured to:

[0127] The fourth data is obtained through the generator of the DCGAN;

[0128] The fourth data and the target structured data are processed by the discriminator of the DCGAN to obtain the fifth data, wherein the target structured data is the structured data corresponding to the target label; the proportion of the target structured data in the total amount of structured data is less than the proportion of other structured data under the target label in the total amount of structured data; in the discriminator, the structured indicators of the fourth data and the target structured data are mapped to different spaces of the image;

[0129] Calculate the loss function for the fourth and fifth data;

[0130] The generator and discriminator are iteratively trained according to the loss function to obtain a data balancing model, and the output of the trained data balancing model is used as the supplementary sample data. The processor is further configured to:

[0131] The processor is further configured to:

[0132] Randomly generate noise data 1×H, where H represents the dimension of the noise data, and the dimension of the noise data is smaller than the dimension of the structured data;

[0133] The noise data is processed by a generator to obtain a fourth data point with 1×N dimensions.

[0134] The processor is further configured to:

[0135] The noise data is input into the fully connected layer of the generator, into a 1×N dimensional third data, where N represents the number of indicators in the structured data, which is an integer greater than 0;

[0136] The 1×N dimensional third data is input into the deconvolution layer of the generator to obtain M×M dimensional data;

[0137] The M×M dimensional data is input into the reshaping layer and clipping layer of the generator to obtain the 1×N dimensional fourth data, where M is... Round up.

[0138] The processor is further configured to:

[0139] The 1×N dimensional fourth data and the target structured data are processed by a discriminator to obtain the 1×N dimensional fifth data.

[0140] The processor is further configured to:

[0141] The 1×N dimensional fourth data and the target structured data are input into the discriminator and mapped into an M×M dimensional target image;

[0142] The spatial characteristics of the target image are extracted using the convolutional layer of the discriminator;

[0143] The output of the convolutional layer of the discriminator is passed through the fully connected layer of the discriminator to obtain the 1×N dimensional fifth data.

[0144] The processor is further configured to:

[0145] Obtain indicator classification information;

[0146] Based on the index classification information, the N indicators of the 1×N dimensional fourth data are mapped to different spaces of the image to obtain an M×M dimensional target image.

[0147] The index classification information is obtained by clustering the structured indicators of the structured data.

[0148] The processor is further configured to:

[0149] The structured indicators corresponding to the structured data are clustered to obtain n clustering indicators, and the information of the n clustering indicators is used as the indicator classification information.

[0150] Where n is an integer and n≥2.

[0151] The processor is further configured to:

[0152] The video quality monitoring model is trained using the target structured data and the supplementary sample data.

[0153] The processor is further configured to:

[0154] The QoE level is obtained by using third-party data obtained from network-side data sources and a trained video quality monitoring model.

[0155] Fourthly, embodiments of this application also provide a communication device, including: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the data processing method described above.

[0156] Fifthly, embodiments of this application also provide a readable storage medium storing a program that, when executed by a processor, implements the steps in the data processing method described above.

[0157] In this embodiment, the acquired structured data is used to generate supplementary sample data through an improved DCGAN. This supplementary sample data is generated based on the mapping results from the target information of the structured data to different spatial representations of the image. The target information includes the clustering results of the structured indicators of the structured data. Because the spatial characteristics of the structured data are utilized to generate the supplementary sample data in the above process, the generated supplementary sample data is closer to the original sample data. This solves the problem of poor model generalization ability caused by a lack of high-quality samples, thereby making the obtained video quality monitoring model more accurate and improving the accuracy of the obtained QoE level. Attached Figure Description

[0158] Figure 1 This is one of the schematic diagrams of the data processing apparatus provided in the embodiments of this application;

[0159] Figure 2 This is a flowchart of the data processing method provided in the embodiments of this application;

[0160] Figure 3 This is a schematic diagram of the structure of DCGAN provided in the embodiments of this application;

[0161] Figure 4 This is a second schematic diagram of the data processing apparatus provided in the embodiments of this application;

[0162] Figure 5 This is the third schematic diagram of the data processing device provided in the embodiments of this application. Detailed Implementation

[0163] In the embodiments of this application, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0164] In the embodiments of this application, the term "multiple" refers to two or more, and other quantifiers are similar.

[0165] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0166] In this embodiment of the application, in order to address the problem of inaccurate evaluation and insufficient generalization ability of AI methods in video experience evaluation due to the small number of poor-quality samples, a method is proposed to cluster structured data composed of existing network indicators according to their correlation and map them onto different spaces of the image. The improved DCGAN is used to extract spatial characteristics, thereby more accurately restoring the real data and generating poor-quality data that is close to the real distribution. By adding the generated data to the samples for model training, the problem of poor model generalization ability and inaccurate evaluation caused by the small number of poor-quality samples can be solved.

[0167] like Figure 1 The diagram shown is a schematic representation of a data processing apparatus according to an embodiment of this application. The apparatus may include: a data collection and storage module 101, a data preprocessing and association module 102, a data balancing processing module 103, and a model training module 104.

[0168] The data collection and storage module 101 is used to acquire data from the video terminal testing tool and network-side data sources, and to preprocess the data, such as cleaning, outlier removal, and missing value imputation. Then, the data acquired from the video terminal testing tool and the data acquired from the network-side data sources are matched.

[0169] The video terminal testing tool monitors the video playback process of end users in real time. By analyzing the video services on the terminal side, it obtains video experience metrics such as initial latency, percentage of buffering time, number of buffering events, and buffering duration at the single playback granularity. These metrics can be used as training labels. Simultaneously, the tool also acquires primary correlation metrics, including key indicators such as user IP, video ID or URL, video playback start time, and video playback end time. Due to limited storage space, the tool automatically pushes local data to the data collection and storage module. If data collection requirements change, the module must send a list of modified fields. The testing tool then sends data to the module according to the revised field list.

[0170] The network-side data source provides network-side data, which can be network elements or systems with network data acquisition and storage capabilities, such as DPI (Deep Packet Inspection) systems, network management systems, and data centers. The data collection and storage module can collect and store data from the network-side data source and video terminal testing tools according to a list. The data obtained from the network-side data source includes video-related metrics and secondary related metrics. The video-related metrics include, but are not limited to, TCP (Transmission Control Protocol) RTT (Round-Trip Time) related metrics; TCP transmission traffic related metrics; out-of-order related metrics; retransmission related metrics; streaming media latency related metrics; and TCP window and configuration related metrics.

[0171] The data balancing processing module 103, based on an improved DCGAN algorithm, trains a data balancing processing model. It can generate samples close to real data from small sample data for training the video experience monitoring model, thereby improving the model's generalization ability. Here, the small sample data refers to data obtained from the matching data that constitutes a small portion of the total matching data.

[0172] The model training module 104 inputs the data generated in the data balancing module and the matching data into the model training module. Labeled by the initial buffering latency and the percentage of stuttering time, it trains a classification model based on xgboost (eXtreme Gradient Boosting) and outputs an executable model. The trained video quality monitoring model is then integrated into the network-side data source for video experience result inference.

[0173] See Figure 2 , Figure 2 This is a flowchart of the data processing method provided in the embodiments of this application, such as... Figure 2 As shown, it includes the following steps:

[0174] Step 201: Obtain structured data.

[0175] The structured data can be structured data directly obtained from other devices or systems, or it can be structured data generated in the following manner.

[0176] In embodiments of this application, data to be processed can be acquired first, comprising: first data from a video terminal testing tool and second data from a network-side data source. The first data includes video experience measurement metrics and a first correlation metric; the second data includes video-related metrics and a second correlation metric. Then, matching data is obtained from the data to be processed, and this matching data is used as the structured data.

[0177] The video terminal testing tool monitors the video playback process of end users in real time. By analyzing the video services on the terminal side, it obtains video experience metrics such as initial buffering latency, percentage of buffering time, number of buffering events, and buffering duration at the granularity of a single playback session. These video experience metrics can be used as training labels. Simultaneously, the video terminal testing tool also obtains primary correlation metrics, including key indicators such as user IP, video ID or URL, video playback start time, and video playback end time. Accordingly, the above data can be obtained here, among which the video experience metrics and primary correlation metrics can be used as primary data.

[0178] Among them, the video playback start time of the first related indicator refers to the moment when the terminal requests to play the video, and the video playback end time refers to the moment when the terminal ends playback.

[0179] The network-side data source can obtain video-related metrics according to a list. This list includes, but is not limited to, one or more of the following video-related metrics: TCP RTT-related metrics; TCP transmission traffic-related metrics; out-of-order metrics; retransmission-related metrics; streaming media latency-related metrics; and TCP window and configuration-related metrics, as shown in Table 1.

[0180] Table 1

[0181]

[0182]

[0183]

[0184] The network-side data source can also obtain secondary related metrics, including key indicators such as user IP, video ID or URL, video playback start time, and video playback end time. Accordingly, the above data can be obtained here, with video-related metrics and secondary related metrics serving as the secondary data.

[0185] Among them, the second related indicator, video playback start time, refers to the moment when the network side starts transmitting video according to the terminal's request, and video playback end time refers to the moment when video transmission ends.

[0186] During the above process, if it is necessary to update the data collection list on the terminal side, the updated list can be sent to the video terminal testing tool, and the video terminal testing tool will collect data according to the updated list.

[0187] Specifically, in obtaining matching data from the data to be processed, the video experience measurement indicators and the video-related indicators are preprocessed to obtain target first data and target second data. The preprocessing may include operations such as data cleaning, outlier removal, missing value imputation, and video ID extraction. The video ID can be extracted from the URL and uniquely identifies the content played by the user; it can be obtained simultaneously from data sent by the video terminal testing tool and data sent by the network-side data source. Then, the target first data and the target second data are matched using a four-tuple to obtain matching data; the four-tuple includes the user IP, video ID, video playback start time difference, and video playback end time difference.

[0188] The video playback start time difference refers to the difference in the start time of video playback for different related indicators, and the video playback end time difference refers to the difference in the end time of video playback for different related indicators. The video playback start time difference and the video playback end time difference can be set as needed.

[0189] When using the quadruple to obtain matching data, the target first data and the target second data can be matched based on the user IP and video ID in the first and second association indicators, respectively. First intermediate data and second intermediate data are obtained from the target first data, where the first and second intermediate data correspond to the same user IP and video ID. Next, based on the video playback start time and video playback end time in the first association indicator corresponding to the first intermediate data and the second association indicator corresponding to the second terminal data, the first intermediate data and the second intermediate data are matched, resulting in third intermediate data and fourth intermediate data. The difference between the video playback start times of the third and fourth intermediate data must meet the requirement for a video playback start time difference, and the difference between the video playback end times of the third and fourth intermediate data must also meet the requirement for a video playback end time difference. Finally, the third and fourth intermediate data are used as the matching data.

[0190] In practical applications, there can be multiple sets of first and second data. Therefore, by matching the user IP and video ID of the first and second data, the first and second intermediate data can be filtered out from the data to be processed. Then, based on the requirements of the video playback start time difference and video playback end time difference, the third and fourth intermediate data are obtained from the first and second intermediate data.

[0191] The difference between the video playback start time of the third intermediate data and the fourth intermediate data meeting the requirements of the video playback start time difference means that the difference between the video playback start time of the third intermediate data and the fourth intermediate data is within the range specified by the video playback start time difference; the difference between the video playback end time of the third intermediate data and the fourth intermediate data meeting the requirements of the video playback end time difference means that the difference between the video playback end time of the third intermediate data and the fourth intermediate data is within the range specified by the video playback end time difference.

[0192] By using a four-tuple matching method, accurate matching and association of data from the video testing terminal and data from the network-side data source can be achieved. The user IP, video ID, and video playback start time difference are used for association, while the video playback end time difference is used to filter and verify in the event of a non-one-to-one match and ensure accurate one-to-one matching of data on both sides.

[0193] Step 202: Generate supplementary sample data based on the improved DCGAN and the structured data. The supplementary sample data is generated based on the mapping results from the target information of the structured data to different spatial representations of the image. The target information includes the clustering results of the structured indicators of the structured data.

[0194] In this embodiment, the mapping of the target information of the structured data to different spaces of the image is performed in the following manner:

[0195] The structured indicators of the structured data are clustered to obtain L types of structured indicators. Then, based on the correspondence between the structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L types of structured indicators is obtained. Then, each type of structured indicator is mapped to the image according to the pixel range corresponding to each type of structured indicator; where L is an integer greater than or equal to 1.

[0196] Specifically, in this step, the fourth data is obtained through the generator of the DCGAN; the fourth data and the target structured data are then processed by the discriminator of the DCGAN to obtain the fifth data, wherein the target structured data is the structured data corresponding to the target label; the proportion of the target structured data in the total amount of structured data is less than the proportion of other structured data under the target label in the total amount of structured data; in the discriminator, the structured indicators of the fourth data and the target structured data are mapped to different spaces of the image. Then, the loss function of the fourth data and the fifth data is calculated, and the generator and the discriminator are iteratively trained repeatedly based on the loss function to obtain a data balancing processing model, and the output of the trained data balancing processing model is used as the supplementary sample data. The preset value can be set according to actual needs.

[0197] Here, the distribution ratio of statistical data can be statistically analyzed according to a predefined data classification method, and the category with the smaller proportion can be extracted from the structured data as the target structured data. As mentioned earlier, each indicator in the video experience measurement metrics obtained by the video terminal testing tool can be used as a label to label the second data from the network-side data source that matches the first data from the video terminal testing tool. Since there are multiple indicators, the second data from the network-side data source that matches the first data from the video terminal testing tool can also have multiple labels. Therefore, what is obtained here is the target structured data under a certain label.

[0198] In this embodiment of the application, supplementary sample data is generated and subsequent model training is performed based on the improved GANN algorithm.

[0199] like Figure 3 As shown, the improved GANN includes a generator G (301) and a discriminator D (302). The generator comprises fully connected layers, deconvolutional layers, reshape layers, and cut layers. The purpose of the fully connected layers is to increase the dimension of the random data to the pre-generated data dimension before inputting it into the deconvolutional layer. The output of the deconvolutional layer is then converted back to the original data dimension. The discriminator includes reshape layers, convolutional layers, and fully connected layers.

[0200] The data generated by the generator and the real data (i.e., the aforementioned target data) are input into the discriminator. According to the data partitioning rules, the data is transformed into an image composed of n large blocks (where n is an integer representing the number of indicators in the matching data) corresponding to different attributes. This image is then input into a convolutional network. The convolutional network extracts the spatial characteristics of the image, and finally, a fully connected layer reduces unread data. Afterward, the loss function between the generated image and the real data is calculated. The generator continuously updates and optimizes its model based on the feedback from the discriminator's loss function, thereby generating samples that more closely resemble the real data. The advantage of this design is that it can accurately extract spatial characteristics and more accurately reproduce the real data.

[0201] Specifically, in this step, obtaining the fourth data through the DCGAN generator may include the following process:

[0202] S1. Randomly generate noise data 1×H, where H represents the dimension of the noise data, and the dimension of the noise data is smaller than the dimension of the matching data. The number of fully connected layers m in the generator is set according to the actual effect.

[0203] S2. The noise data is processed by a generator to obtain the first data in 1×N dimensions.

[0204] Specifically, in this step, the noise data is input into the fully connected layer of the generator to obtain 1×N dimensional third data, where N represents the number of metrics in the matching data and is a positive integer. Then, the 1×N dimensional third data is input into the deconvolution layer of the generator to obtain M×M dimensional data; where M is... Round up. Finally, the M×M dimensional data is input into the generator's reshaping and clipping layers to obtain the 1×N dimensional first data. After the reshaping layer, the 1×M dimensional data is obtained. 2 The image, after being cropped, yields the first data in 1×N dimensions.

[0205] Specifically, in this step, the fourth data and the target structured data are processed by the discriminator of the DCGAN to obtain the fifth data, including:

[0206] S1. The 1×N dimensional first data and the target data are processed by a discriminator to obtain 1×N dimensional second data.

[0207] Specifically, this step may include:

[0208] S11. Input the 1×N dimensional first data and the target data into the discriminator to map them into an M×M dimensional target image.

[0209] Specifically, in this step, indicator classification information can be obtained first. Then, based on the indicator classification information, the reshaping layer of the discriminator maps the N indicators of the 1×N-dimensional first data to different spaces of the image, thereby obtaining an M×M-dimensional target image.

[0210] The indicator classification information is obtained by clustering the indicators included in the matching data.

[0211] Specifically, the indicators included in the matching data can be clustered, dividing these structural indicators into n categories based on their correlation. According to this classification information, the n clustered indicators can be mapped to different spatial regions of the image in the discriminator's reshaping layer, transforming it into an image composed of n large blocks of indicators with different attributes. For example, during the image transformation process, different categories of indicators can be set to different pixel ranges. Therefore, for one or more indicators within a certain category, different pixel values ​​can be set within the corresponding pixel range. In this way, the obtained image contains spatial information, and spatially adjacent pixels have a very strong correlation.

[0212] S12. Extract the spatial characteristics of the target image using the convolutional layer of the discriminator.

[0213] S13. The output of the convolutional layer of the discriminator is passed through the fully connected layer of the discriminator to obtain the 1×N-dimensional fifth data.

[0214] In practical applications, the first data generated by the generator and the target data (1×N dimensions) can be simultaneously input into the discriminator. To further improve accuracy, the input data can first be padded with 0s. Then, the indicators in the input data can be mapped into an M×M dimensional image according to the rules for obtaining indicator classification information. This image is then output after passing through convolutional layers and fully connected layers.

[0215] In this step, the loss function can be calculated using the following formula:

[0216]

[0217] In the above formula, G represents the generator, z represents random noise, and the data generated by the generator is denoted as G(z). D represents the discriminator network, x represents the target data, D(x) represents the probability that x is the target data, and D(G(z)) represents the probability that the data generated by the generator is true. data p represents the target data distribution. z V represents the noise distribution defined in a low dimension; V represents the value function; min GV(D, G) represents the training of the generator to minimize the probability that the data generated by the generator is identified as fake data when input into the discriminator; max D V(D,G) represents maximizing the sum of the probability that the discriminator makes the data generated by the discriminator generator true and the probability that the data generated by the discriminator generator is false.

[0218] In this embodiment, the acquired structured data is used to generate supplementary sample data through an improved DCGAN. This supplementary sample data is generated based on the mapping results from the target information of the structured data to different spatial representations of the image. The target information includes the clustering results of the structured indicators of the structured data. Because the spatial characteristics of the structured data are utilized to generate the supplementary sample data in the above process, the generated supplementary sample data is closer to the original sample data. This solves the problem of poor model generalization ability caused by a lack of high-quality samples, thereby making the obtained video quality monitoring model more accurate and improving the accuracy of the obtained QoE level.

[0219] Based on the above embodiments, the method may further include:

[0220] The video quality monitoring model is trained using the matching data and the supplementary sample data.

[0221] The matching data and the supplementary sample data are mixed and randomly sorted. Then, the initial delay and the percentage of stuttering are used as labels. A classification model is trained based on machine learning or deep learning to output a video quality monitoring model.

[0222] Based on the above embodiments, the Quality of Experience (QoE) level can also be obtained using third data obtained from the network-side data source and a trained video quality monitoring model. The third data may include video-related indicators, or further, correlation indicators. The third data may also be the second data described above.

[0223] In practical applications, an interface can be added to the network-side data source to integrate the trained video quality monitoring model into the network-side data source for result inference.

[0224] Based on video-related metrics obtained from network-side data sources, an integrated video quality monitoring model is used to output the QoE level at the user level to help operations and maintenance personnel identify quality issues.

[0225] This application also provides a data processing apparatus. See [link to previous document]. Figure 4 , Figure 4This is a structural diagram of the data processing apparatus provided in the embodiments of this application. Since the principle by which the data processing apparatus solves the problem is similar to the data processing method in the embodiments of this application, the implementation of this data processing apparatus can refer to the implementation of the method, and repeated details will not be described again.

[0226] like Figure 4 As shown, the data processing device 400 includes:

[0227] The first acquisition module 401 is used to acquire structured data; the first generation module 402 is used to generate supplementary sample data based on the improved DCGAN and the structured data.

[0228] The supplementary sample data is generated based on the mapping results of the target information of the structured data to different spaces of the image, and the target information includes the clustering results of the structured indicators of the structured data.

[0229] The first acquisition module includes:

[0230] The first acquisition submodule is used to acquire data to be processed, wherein the data to be processed includes: first data from the video terminal testing tool and second data from the network-side data source. The first data includes video experience measurement indicators and first related indicators; the second data includes video-related indicators and second related indicators.

[0231] The second acquisition submodule is used to obtain matching data from the data to be processed and use the matching data as the structured data.

[0232] The relevant metrics include: user IP, video ID or URL, video playback start time, and video playback end time.

[0233] The second acquisition submodule includes:

[0234] The preprocessing unit is used to preprocess the first data and the second data to obtain target first data and target second data;

[0235] A matching unit is used to match the first target data and the second target data using a quadruple to obtain matching data;

[0236] The quadruple includes the user IP, video ID, video playback start time difference, and video playback end time difference.

[0237] The matching unit includes:

[0238] The first matching subunit is configured to match the target first data and the target second data according to the user IP and video ID in the first association index and the second association index, obtain first intermediate data from the target first data and second intermediate data from the target second data, wherein the first intermediate data and the second intermediate data correspond to the same user IP and video ID;

[0239] The second matching subunit is used to match the first intermediate data and the second intermediate data according to the video playback start time and video playback end time in the first correlation index corresponding to the first intermediate data and the video playback start time and video playback end time in the second correlation index corresponding to the second intermediate data, to obtain the third intermediate data from the target first data and the fourth intermediate data from the target second data, wherein the difference between the video playback start time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback start time difference, and the difference between the video playback end time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback end time difference;

[0240] The first acquisition subunit is used to use the third intermediate data and the fourth intermediate data as the matching data.

[0241] Specifically, the mapping of the target information of the structured data to different spaces of the image is performed in the following manner:

[0242] The structured indicators of the structured data are clustered to obtain L types of structured indicators;

[0243] Based on the correspondence between structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L-type structured indicators is obtained.

[0244] Each type of structured index is mapped to the image based on the pixel range corresponding to it; where L is an integer greater than or equal to 1.

[0245] The first generation module includes:

[0246] The first acquisition submodule is used to obtain the fourth data through the generator of the DCGAN;

[0247] The second acquisition submodule is used to process the fourth data and the target structured data through the discriminator of the DCGAN to obtain the fifth data, wherein the target structured data is the structured data corresponding to the target label; the proportion of the target structured data in the total amount of structured data is less than the proportion of other structured data under the target label in the total amount of structured data; in the discriminator, the structured indicators of the fourth data and the target structured data are mapped to different spaces of the image;

[0248] The first calculation submodule is used to calculate the loss function of the fourth data and the fifth data;

[0249] The third acquisition submodule is used to repeatedly iterate and train the generator and the discriminator according to the loss function to obtain the data balance processing model, and use the output of the trained data balance processing model as the supplementary sample data.

[0250] The first acquisition submodule includes:

[0251] The first generation unit is used to randomly generate noise data 1×H, where H represents the dimension of the noise data, and the dimension of the noise data is smaller than the dimension of the structured data.

[0252] The second generation unit is used to process the noise data through a generator to obtain 1×N dimensional fourth data.

[0253] The second generation unit includes:

[0254] The first generation subunit is used to input the noise data into the fully connected layer of the generator, to the 1×N dimensional third data, where N represents the number of indicators of the structured data, which is an integer greater than 0;

[0255] The second generation subunit is used to input the 1×N dimensional third data into the deconvolution layer of the generator to obtain M×M dimensional data;

[0256] The third generation subunit is used to input the M×M dimensional data into the reshaping layer and clipping layer of the generator to obtain the 1×N dimensional fourth data, where M is... Round up.

[0257] The second acquisition submodule is used to pass the 1×N-dimensional fourth data and the target structured data through a discriminator to obtain 1×N-dimensional fifth data.

[0258] The second acquisition submodule includes:

[0259] The first generation unit is used to input the 1×N dimensional fourth data and the target structured data into the discriminator and map them into an M×M dimensional target image.

[0260] The first extraction unit is used to extract the spatial characteristics of the target image using the convolutional layer of the discriminator;

[0261] The second generation unit is used to obtain the 1×N-dimensional fifth data by passing the output of the convolutional layer of the discriminator through the fully connected layer of the discriminator.

[0262] The first generation unit includes:

[0263] The first acquisition subunit is used to acquire indicator classification information;

[0264] The first mapping subunit is used to map the N indicators of the 1×N-dimensional fourth data to different spaces of the image according to the indicator classification information, so as to obtain an M×M-dimensional target image.

[0265] The index classification information is obtained by clustering the structured indicators of the structured data.

[0266] The first acquisition subunit is used to cluster the structured indicators corresponding to the structured data to obtain n clustering indicators, and use the information of the n clustering indicators as the indicator classification information.

[0267] Where n is an integer and n≥2.

[0268] The device further includes:

[0269] The training module is used to train the video quality monitoring model using the target structured data and the supplementary sample data.

[0270] The device further includes:

[0271] The second acquisition module is used to obtain the QoE level by using third data obtained from the network-side data source and the trained video quality monitoring model.

[0272] The apparatus provided in this application embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0273] This application also provides a data processing apparatus. See [link to previous document]. Figure 5 , Figure 5 This is a structural diagram of the data processing apparatus provided in the embodiments of this application. Since the principle by which the data processing apparatus solves the problem is similar to the data processing method in the embodiments of this application, the implementation of this data processing apparatus can refer to the implementation of the method, and repeated details will not be described again.

[0274] like Figure 5 As shown, the data processing device includes a processor 501 and a transceiver 502; wherein, the processor 501 is used for:

[0275] Obtain structured data;

[0276] Supplementary sample data is generated based on the improved DCGAN and the structured data;

[0277] The supplementary sample data is generated based on the mapping results of the target information of the structured data to different spaces of the image, and the target information includes the clustering results of the structured indicators of the structured data.

[0278] The processor is further configured to:

[0279] Acquire data to be processed, wherein the data to be processed includes: first data from a video terminal testing tool and second data from a network-side data source, wherein the first data includes video experience measurement indicators and a first related indicator; and the second data includes video-related indicators and a second related indicator.

[0280] Matching data is obtained from the data to be processed, and the matching data is used as the structured data.

[0281] The associated metrics include: user IP, video ID or URL, video playback start time, and video playback end time; the processor is also used for:

[0282] Preprocessing the first data and the second data yields the target first data and the target second data;

[0283] The target first data and the target second data are matched using a quadruple to obtain matching data;

[0284] The quadruple includes the user IP, video ID, video playback start time difference, and video playback end time difference.

[0285] The processor is further configured to:

[0286] Based on the user IP and video ID in the first and second association indicators, the target first data and the target second data are matched to obtain first intermediate data from the target first data and second intermediate data from the target second data, wherein the first intermediate data and the second intermediate data correspond to the same user IP and video ID;

[0287] Based on the video playback start time and video playback end time in the first correlation index corresponding to the first intermediate data and the video playback start time and video playback end time in the second correlation index corresponding to the second intermediate data, the first intermediate data and the second intermediate data are matched to obtain third intermediate data from the target first data and fourth intermediate data from the target second data. The difference between the video playback start time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback start time difference, and the difference between the video playback end time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback end time difference.

[0288] The third intermediate data and the fourth intermediate data are used as the matching data.

[0289] The processor is further configured to: map the target information of the structured data to different spaces of the image in the following manner:

[0290] The structured indicators of the structured data are clustered to obtain L types of structured indicators;

[0291] Based on the correspondence between structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L-type structured indicators is obtained.

[0292] Each type of structured index is mapped to the image based on the pixel range corresponding to it; where L is an integer greater than or equal to 1.

[0293] The processor is further configured to:

[0294] The fourth data is obtained through the generator of the DCGAN;

[0295] The fourth data and the target structured data are processed by the discriminator of the DCGAN to obtain the fifth data, wherein the target structured data is the structured data corresponding to the target label; the proportion of the target structured data in the total amount of structured data is less than the proportion of other structured data under the target label in the total amount of structured data; in the discriminator, the structured indicators of the fourth data and the target structured data are mapped to different spaces of the image;

[0296] Calculate the loss function for the fourth and fifth data;

[0297] The generator and the discriminator are repeatedly trained iteratively according to the loss function to obtain a data balancing model, and the output of the trained data balancing model is used as the supplementary sample data.

[0298] The processor is further configured to:

[0299] Randomly generate noise data 1×H, where H represents the dimension of the noise data, and the dimension of the noise data is smaller than the dimension of the structured data;

[0300] The noise data is processed by a generator to obtain a fourth data point with 1×N dimensions.

[0301] The processor is further configured to:

[0302] The noise data is input into the fully connected layer of the generator, into a 1×N dimensional third data, where N represents the number of indicators in the structured data, which is an integer greater than 0;

[0303] The 1×N dimensional third data is input into the deconvolution layer of the generator to obtain M×M dimensional data;

[0304] The M×M dimensional data is input into the reshaping layer and clipping layer of the generator to obtain the 1×N dimensional fourth data, where M is... Round up.

[0305] The processor is further configured to:

[0306] The 1×N dimensional fourth data and the target structured data are processed by a discriminator to obtain the 1×N dimensional fifth data.

[0307] The processor is further configured to:

[0308] The 1×N dimensional fourth data and the target structured data are input into the discriminator and mapped into an M×M dimensional target image;

[0309] The spatial characteristics of the target image are extracted using the convolutional layer of the discriminator;

[0310] The output of the convolutional layer of the discriminator is passed through the fully connected layer of the discriminator to obtain the 1×N dimensional fifth data.

[0311] The processor is further configured to:

[0312] Obtain indicator classification information;

[0313] Based on the index classification information, the N indicators of the 1×N dimensional fourth data are mapped to different spaces of the image to obtain an M×M dimensional target image.

[0314] The index classification information is obtained by clustering the structured indicators of the structured data.

[0315] The processor is further configured to:

[0316] The structured indicators corresponding to the structured data are clustered to obtain n clustering indicators, and the information of the n clustering indicators is used as the indicator classification information.

[0317] Where n is an integer and n≥2.

[0318] The processor is further configured to:

[0319] The video quality monitoring model is trained using the target structured data and the supplementary sample data.

[0320] The processor is further configured to:

[0321] The QoE level is obtained by using third-party data obtained from network-side data sources and a trained video quality monitoring model.

[0322] The data to be processed is obtained at the granularity of a single video playback by a user.

[0323] The apparatus provided in this application embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0324] It should be noted that the division of units in the embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0325] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0326] This application also provides a communication device, including: a transceiver, a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the data processing method described above.

[0327] This application also provides a readable storage medium storing a program. When executed by a processor, this program implements the various processes of the above-described data processing method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical storage (e.g., CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0328] It should be noted that, in this document, 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 a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0329] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0330] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A data processing method, characterized in that, include: Obtain structured data; Supplementary sample data is generated based on the improved deep convolutional generative adversarial network (DCGAN) and the structured data; the improved deep convolutional generative adversarial network (DCGAN) includes a generator and a judge. The supplementary sample data is generated based on the mapping results of the target information of the structured data to different spaces of the image, and the target information includes the clustering results of the structured indicators of the structured data; Specifically, the mapping of the target information of the structured data to different spaces of the image is performed in the following manner: The structured indicators of the structured data are clustered to obtain L types of structured indicators; Based on the correspondence between structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L-type structured indicators is obtained. Based on the pixel range corresponding to each type of structured indicator, map each type of structured indicator to the image; Where L is an integer greater than or equal to 1.

2. The method according to claim 1, characterized in that, The acquisition of structured data includes: Acquire data to be processed, wherein the data to be processed includes: first data from a video terminal testing tool and second data from a network-side data source, wherein the first data includes video experience measurement indicators and a first related indicator; and the second data includes video-related indicators and a second related indicator. Matching data is obtained from the data to be processed, and the matching data is used as the structured data.

3. The method according to claim 2, characterized in that, Related metrics include: user IP, video ID or Uniform Resource Locator URL, video playback start time, and video playback end time; Obtaining matching data from the data to be processed includes: Preprocessing the first data and the second data yields the target first data and the target second data; The target first data and the target second data are matched using a quadruple to obtain matching data; The quadruple includes the user IP, video ID, video playback start time difference, and video playback end time difference.

4. The method according to claim 3, characterized in that, The process of matching the first target data and the second target data using quadruples to obtain matching data includes: Based on the user IP and video ID in the first and second association indicators, the target first data and the target second data are matched to obtain first intermediate data from the target first data and second intermediate data from the target second data, wherein the first intermediate data and the second intermediate data correspond to the same user IP and video ID; Based on the video playback start time and video playback end time in the first correlation index corresponding to the first intermediate data and the video playback start time and video playback end time in the second correlation index corresponding to the second intermediate data, the first intermediate data and the second intermediate data are matched to obtain third intermediate data from the target first data and fourth intermediate data from the target second data. The difference between the video playback start time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback start time difference, and the difference between the video playback end time of the third intermediate data and the fourth intermediate data meets the requirement of the video playback end time difference. The third intermediate data and the fourth intermediate data are used as the matching data.

5. The method according to claim 1, characterized in that, The process of generating supplementary sample data based on the improved deep convolutional generative adversarial network (DCGAN) and the structured data includes: The fourth data is obtained through the generator of the DCGAN; The fourth data and the target structured data are processed by the discriminator of the DCGAN to obtain the fifth data, wherein the target structured data is the structured data corresponding to the target label; the proportion of the target structured data in the total amount of structured data is less than the proportion of other structured data under the target label in the total amount of structured data; the structured indicators of the fourth data and the target structured data are mapped to different spaces of the image; Calculate the loss function for the fourth and fifth data; The generator and the discriminator are repeatedly trained iteratively according to the loss function to obtain a data balancing model, and the output of the trained data balancing model is used as the supplementary sample data.

6. The method according to claim 5, characterized in that, The process of obtaining the fourth data through the generator of the DCGAN includes: Randomly generate noise data 1×H, where H represents the dimension of the noise data, and the dimension of the noise data is smaller than the dimension of the structured data; The noise data is processed by a generator to obtain a fourth data point with 1×N dimensions.

7. The method according to claim 6, characterized in that, The step of passing the noise data through a generator to obtain 1×N dimensional fourth data includes: The noise data is input into the fully connected layer of the generator to obtain 1×N dimensional third data, where N represents the number of indicators in the structured data and N is an integer greater than 0. The 1×N dimensional third data is input into the deconvolution layer of the generator to obtain M×M dimensional data; The M×M dimensional data is input into the reshaping layer and clipping layer of the generator to obtain the 1×N dimensional fourth data, where M is... Round up.

8. The method according to claim 5, characterized in that, The step of passing the fourth data and the target structured data through the discriminator of the DCGAN to obtain the fifth data includes: The 1×N dimensional fourth data and the target structured data are processed by a discriminator to obtain the 1×N dimensional fifth data.

9. The method according to claim 8, characterized in that, The step of passing the 1×N dimensional fourth data and the target structured data through a discriminator to obtain the 1×N dimensional fifth data includes: The 1×N dimensional fourth data and the target structured data are input into the discriminator and mapped into an M×M dimensional target image; The spatial characteristics of the target image are extracted using the convolutional layer of the discriminator; The output of the convolutional layer of the discriminator is passed through the fully connected layer of the discriminator to obtain the 1×N dimensional fifth data.

10. The method according to claim 9, characterized in that, The step of inputting the 1×N dimensional fourth data and the target structured data into the discriminator and mapping them into an M×M dimensional target image includes: Obtain indicator classification information; Based on the index classification information, the N indicators of the 1×N dimensional fourth data are mapped to different spaces of the image to obtain an M×M dimensional target image. The index classification information is obtained by clustering the structured indicators of the structured data.

11. The method according to claim 10, characterized in that, The acquisition of indicator classification information includes: The structured indicators corresponding to the structured data are clustered to obtain n clustering indicators, and the information of the n clustering indicators is used as the indicator classification information. Where n is an integer and n≥2.

12. The method according to claim 5, characterized in that, After generating the supplementary sample data, the method further includes: The video quality monitoring model is trained using the target structured data and the supplementary sample data.

13. The method according to claim 12, characterized in that, After training the video quality monitoring model, the method further includes: The Quality of Experience (QoE) level is obtained by using third-party data obtained from network-side data sources and a trained video quality monitoring model.

14. A data processing apparatus, characterized in that, include: The first acquisition module is used to acquire structured data; The first generation module is used to generate supplementary sample data based on the improved DCGAN and the structured data; the improved deep convolutional generative adversarial network DCGAN includes a generator and a judge. The supplementary sample data is generated based on the mapping results of the target information of the structured data to different spaces of the image, and the target information includes the clustering results of the structured indicators of the structured data; Specifically, the mapping of the target information of the structured data to different spaces of the image is performed in the following manner: The structured indicators of the structured data are clustered to obtain L types of structured indicators; Based on the correspondence between structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L-type structured indicators is obtained. Based on the pixel range corresponding to each type of structured indicator, map each type of structured indicator to the image; Where L is an integer greater than or equal to 1.

15. A data processing apparatus, characterized in that, include: Processor and transceiver; wherein the processor is used for: Obtain structured data; Supplementary sample data is generated based on the improved DCGAN and the structured data; the improved deep convolutional generative adversarial network DCGAN includes a generator and a judge. The supplementary sample data is generated based on the mapping results of the target information of the structured data to different spaces of the image, and the target information includes the clustering results of the structured indicators of the structured data; Specifically, the mapping of the target information of the structured data to different spaces of the image is performed in the following manner: The structured indicators of the structured data are clustered to obtain L types of structured indicators; Based on the correspondence between structured indicators and pixel ranges, the pixel range corresponding to each type of structured indicator in the L-type structured indicators is obtained. Based on the pixel range corresponding to each type of structured indicator, map each type of structured indicator to the image; Where L is an integer greater than or equal to 1.

16. A communication device, comprising: A transceiver, a memory, a processor, and a program stored in the memory and executable on the processor; characterized in that the processor is configured to read the program in the memory to implement the steps of the data processing method as described in any one of claims 1 to 13.

17. A readable storage medium for storing a program, characterized in that, When the program is executed by the processor, it implements the steps of the data processing method as described in any one of claims 1 to 13.

Images