Method and device for depicting QoS-QoE mapping relation in audio and video service, and medium

Abstract

The invention provides a method and a device for describing a QoS-QoE mapping relation in an audio and video service, and a medium. The method comprises the following steps: collecting QoS feature data, QoE data and confounding factor data of the audio and video service; fitting the Qos-QoE mapping relation by adopting a machine learning model to obtain a mapping relation fitting model; according to a mapping relation fitting model, screening out confounding factors which have obvious influence on the QoS-QoE mapping relation from the confounding factor data, and screening out QoS main characteristics from the QoS characteristic data; classifying the QoS main characteristic data and the QoE data according to confounding factors with obvious influence to obtain a plurality of data groups, and determining a QoS-QoE quantitative relation curve corresponding to each data group; and clustering the curves of the data groups according to the similarity of the Qos-QoE quantitative relation curves of the data groups to obtain a change mode of the Qos-QoE quantitative relation. According to the method, the influence of hybrid factors is considered, so that the mapping relation analysis is more accurate, and the influence rule of QoS on QoE in different scenes is intuitively summarized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of audio and video content distribution, and in particular to a method and device for characterizing QoS-QoE mapping relationship in audio and video service, and a medium. BACKGROUND

[0002] With the rapid development of Internet video services, the Quality of Experience (QoE) has become a key factor affecting user satisfaction and retention rate. The client evaluates the pros and cons of the service according to the QoE, while the server side cannot obtain the QoE information, so it can only take the Quality of Service (QoS) as the optimization target. There are two kinds of mismatch between QoS and QoE: one is hierarchical mismatch. QoE is the subjective experience of users in the application layer, while QoS is generally implemented by the transport layer component. The other is location mismatch. QoE is obtained by processing the client information, while QoS is monitored and managed by the server. These two mismatch relationships make the relationship between QoS and QoE not a simple linear relationship, so optimizing QoS on the server side does not necessarily improve user experience, and may even make user experience worse. At the same time, the mapping relationship between QoS and QoE is not constant, and time, geographical location and other mixed factors may affect the mapping relationship. Therefore, the same QoS quality may correspond to different QoE qualities at different times and in different access modes. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides a method and device for characterizing QoS-QoE mapping relationship in audio and video service, and a medium.

[0004] In one aspect, the present application provides a method for characterizing QoS-QoE mapping relationship in audio and video service, comprising:

[0005] Collecting QoS feature data, QoE data and mixed factor data of the audio and video service, wherein the QoS feature data is the transport layer performance data in the audio and video transmission process, the QoE data is the subjective experience quality data of the user for the audio and video service, and the mixed factor data is the environment-related data affecting the QoS-QoE mapping relationship;

[0006] Combining the QoS feature data, QoE data and mixed factor data, a machine learning model is used to fit the QoS-QoE mapping relationship to obtain a mapping relationship fitting model;

[0007] fitting a model according to the mapping relationship, screening out a mixed factor having a significant influence on the QoS-QoE mapping relationship from the mixed factor data, and screening out a QoS main feature from the QoS feature data;

[0008] classifying the QoS main feature data and the QoE data according to the mixed factor having the significant influence to obtain a plurality of data groups, and determining a QoS-QoE quantitative relationship curve corresponding to each data group;

[0009] clustering the curves of the data groups according to the similarity of the QoS-QoE quantitative relationship curves of each data group to obtain a change mode of the QoS-QoE quantitative relationship.

[0010] In an embodiment of the present application, the QoS-QoE mapping relationship is fitted by using a machine learning model in combination with the QoS feature data, the QoE data and the mixed factor data, and a mapping relationship fitting model is obtained, which contains:

[0011] The QoS feature data, the mixed factor data and the QoE data are sorted in chronological order;

[0012] The sorted data is divided into a training set and a test set according to a preset proportion;

[0013] The QoS feature data and the mixed factor data in the training set are taken as input, and the QoE data is taken as a label to train the machine learning model, the model fitting error is optimized through hyperparameter adjustment and grid search, and the mapping relationship fitting model is obtained;

[0014] The mapping relationship fitting model is verified by using the test set, and when the error is lower than a preset threshold, it is determined that the mapping relationship fitting model is effective.

[0015] In an embodiment of the present application, the mixed factor having a significant influence on the QoS-QoE mapping relationship is screened out from the mixed factor data by comparing the model fitting errors with and without the target mixed factor, which contains:

[0016] The QoS feature data and all the mixed factor data are taken as input, and the QoE data is taken as a label to train a first mapping model based on the machine learning model;

[0017] After removing the target mixed factor, the QoS feature data and the remaining mixed factor data are taken as input, and the QoE data is taken as a label to train a second mapping model based on the machine learning model;

[0018] If the fitting error of the second mapping model is significantly greater than the fitting error of the first mapping model, it is determined that the target confounding factor is the significantly influential confounding factor.

[0019] In an embodiment of the present application, the QoS main feature is screened from the QoS feature data by comparing the model fitting errors when containing a single QoS feature or all QoS features, including:

[0020] For each QoS feature of the QoS feature data, a third mapping model is trained based on the machine learning model by taking the single QoS feature and the confounding factor data as input and the QoE data as label;

[0021] A fourth mapping model is trained based on the machine learning model by taking all QoS features and the confounding factor data as input and the QoE data as label;

[0022] The fitting errors of the third mapping models and the fourth mapping model are compared, and the QoS feature corresponding to the third mapping model with the fitting error closest to the fourth mapping model is selected as the QoS main feature.

[0023] In an embodiment of the present application, the QoS main feature data and QoE data are classified according to the significantly influential confounding factor to obtain multiple data groups, and the QoS-QoE quantitative relationship curve corresponding to each data group is determined, including:

[0024] The QoS main feature data and QoE data are classified according to the screened significantly influential confounding factor to obtain multiple data groups;

[0025] In each data group, the QoS main feature data is sorted by numerical value and evenly divided into multiple intervals;

[0026] The mean of the QoS main feature data and the mean of the QoE data in each interval are calculated to obtain multiple QoS-QoE mapping points;

[0027] Based on the multiple QoS-QoE mapping points, the QoS-QoE quantitative relationship curve corresponding to the data group is constructed.

[0028] In an embodiment of the present application, the curves of each data group are clustered according to the similarity of the QoS-QoE quantitative relationship curve of each data group, specifically, the similarity between QoS-QoE quantitative relationship curves is calculated by using dynamic time warping algorithm, and then the KMedoids algorithm is used to cluster the curves.

[0029] In an embodiment of the present application, the QoE data and the client-side hybrid factor data of the audio and video service are collected at the client, and the client-side hybrid factor data includes geographical location information data of the client, access mode information data;

[0030] The QoS characteristic data and the server-side hybrid factor data of the audio and video service are collected at the server, and the server-side hybrid factor data includes time information data of the audio and video transmission, geographical location information data of the server, and access operator information data; wherein the hybrid factor data includes the client-side hybrid factor data and the server-side hybrid factor data.

[0031] In an embodiment of the present application, further comprising:

[0032] At the server, the change mode of the QoS-QoE quantitative relationship is used in combination with the real-time QoE data and the client-side hybrid factor data of the current client to determine the matched audio and video transmission QoS parameter, so as to adjust the audio and video data transmission between the client and the server.

[0033] The present application also provides a device for describing the QoS-QoE mapping relationship in an audio and video service, comprising:

[0034] The acquisition module is used to collect the QoS characteristic data, the QoE data and the hybrid factor data of the audio and video service, wherein the QoS characteristic data is the transmission layer performance data in the audio and video transmission process, the QoE data is the subjective experience quality data of the user to the audio and video service, and the hybrid factor data is the environment-related data affecting the QoS-QoE mapping relationship;

[0035] The mapping module is used to combine the QoS characteristic data, the QoE data and the hybrid factor data, and use a machine learning model to fit the QoS-QoE mapping relationship to obtain a mapping relationship fitting model;

[0036] The screening module is used to screen out the hybrid factors having a significant influence on the QoS-QoE mapping relationship from the hybrid factor data according to the mapping relationship fitting model, and screen out the QoS main features from the QoS characteristic data;

[0037] The classification module is used to classify the QoS main feature data and the QoE data according to the hybrid factors having a significant influence, to obtain a plurality of data groups, and determine the QoS-QoE quantitative relationship curve corresponding to each data group;

[0038] The clustering module is used to cluster the curves of the data groups according to the similarity of the QoS-QoE quantitative relationship curve of each data group, to obtain the change mode of the QoS-QoE quantitative relationship.

[0039] In still another aspect, the present application provides a computer readable storage medium, having stored thereon a computer program which, when executed by a processor, implements the steps of the method for characterizing the QoS-QoE mapping relationship in an audio / video service.

[0040] In still another aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the method for characterizing the QoS-QoE mapping relationship in an audio / video service.

[0041] From the above solution, the present application has the following advantages:

[0042] The method for characterizing the QoS-QoE mapping relationship in an audio / video service disclosed in the present application combines a machine learning model to fit the QoS-QoE mapping relationship, breaks the restriction of the mismatch between the hierarchy and position of QoS and QoE in the prior art, and enables the server to optimize QoS based on the real user subjective experience (QoE), thereby avoiding the problem of the decline in user experience caused by the traditional optimization of QoS only. By screening the mixed factors that have a significant impact on the mapping relationship and analyzing the quantitative relationship between QoS and QoE according to the classification, the problem of dynamic changes in the mapping relationship caused by mixed factors in the prior art is solved, the mapping relationship analysis is more accurate, and irrelevant factors are avoided. In addition, by determining the main features of QoS, the difficulty in quantitative analysis caused by the correlation between multiple QoS features is solved, and the QoS-QoE relationship is clearer. By clustering the QoS-QoE quantitative relationship curves under different combinations of mixed factors, the change mode obtained can intuitively summarize the influence law of QoS on QoE in different scenarios, so that the server can match the corresponding mapping mode according to the mixed factors of the current client, dynamically adjust the QoS optimization direction, and adapt to the changes in the mapping relationship in a complex network environment. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 Fig. 1 shows the overall flowchart of the method for characterizing the QoS-QoE mapping relationship in an audio / video service provided by an embodiment of the present application;

[0044] Figure 2 Fig. 2 shows the principle of step S4; Figure 1 Fig. 3 shows an example of array grouping division of step S4;

[0045] Figure 3 Fig. 4 shows the principle diagram of step S4;

[0046] Figure 4 Fig. 5 shows the effect comparison of dynamic time warping matching and Euclidean distance matching;

[0047] Figure 5 Fig. 6 shows the change mode of the QoS-QoE quantitative relationship curve.

[0048] Figure 6 Fig. 1 shows a structural schematic diagram of a device for depicting a QoS-QoE mapping relationship in an audio-video service according to an embodiment of the present application.

[0049] In the drawings:

[0050] 300: a device for depicting a QoS-QoE mapping relationship in an audio-video service;

[0051] 310: an acquisition module;

[0052] 320: a mapping module;

[0053] 330: a screening module;

[0054] 340: a classification module;

[0055] 350: a clustering module. DETAILED DESCRIPTION

[0056] It should be noted that the relationship terms such as first and second and the like in the present application are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0057] Without more limitations, the element defined by the phrase "including a" does not exclude the presence of another identical element in the process, method, article or device including the element.

[0058] Referring to Figure 1 as shown, Figure 1 Fig. 2 shows a general flowchart of a method for depicting a QoS-QoE mapping relationship in an audio-video service according to an embodiment of the present application.

[0059] A method for depicting a QoS-QoE mapping relationship in an audio-video service, comprising:

[0060] Step S1, collecting QoS characteristic data, QoE data and hybrid factor data of the audio-video service, wherein the QoS characteristic data is transmission layer performance data in an audio-video transmission process, the QoE data is subjective experience quality data of a user on the audio-video service, and the hybrid factor data is environment-related data affecting the QoS-QoE mapping relationship.

[0061] In a specific implementation, the QoE data and the client-side hybrid factor data of the audio and video service are collected at the client, and the client-side hybrid factor data includes geographical location information data and access mode information data of the client.

[0062] The QoS characteristic data and the server-side hybrid factor data of the audio and video service are collected at the server, and the server-side hybrid factor data includes time information data of audio and video transmission, geographical location information data of the server, and access operator information data; wherein the hybrid factor data includes the client-side hybrid factor data and the server-side hybrid factor data.

[0063] The QoS characteristic data includes throughput rate data, packet loss rate data, and delay data.

[0064] In step S2, the QoS-QoE mapping relationship is fitted by using a machine learning model in combination with the QoS characteristic data, the QoE data, and the hybrid factor data, to obtain a mapping relationship fitting model.

[0065] In actual application, the influencing factors of QoE are complex and the QoS-QoE mapping relationship may have nonlinear changes. The changes of hybrid factors will cause differences in the mapping relationship in different scenarios, so it is difficult to accurately reflect the QoE quality by only considering QoS. Meanwhile, the existing technology based on the mathematical model is difficult to fit the relatively complex mapping relationship change trend, and there is a large error. In view of this, in an embodiment of the present application, a machine learning model is used to fit the QoS-QoE mapping relationship. At the same time, in order to consider the influence of hybrid factors, the model takes the QoS characteristic data and the hybrid factor data as input at the same time, the QoS characteristic data is a continuous variable, and the hybrid factor data is converted to category coding through feature engineering.

[0066] Specifically, since time as a hybrid factor may affect the mapping relationship, the QoS characteristic data, the hybrid factor data, and the QoE data are first sorted in time sequence. Then, the sorted data is divided into a training set and a test set according to a preset ratio, for example, the preset ratio is 70:30, and the training set and the test set are divided. The QoS characteristic data and the hybrid factor data in the training set are taken as input, and the QoE data is taken as label, the machine learning model is trained, the model fitting error is optimized through hyperparameter adjustment and grid search, and the mapping relationship fitting model is obtained. At the same time, the test set is further used to verify the error of the mapping relationship fitting model, and when the error is lower than a preset threshold, it is determined that the mapping relationship fitting model is effective. If F represents the mapping relationship, the relationship between QoE and each QoS index can be represented as:

[0067]

[0068] , Equation (1)

[0069] In this embodiment, a machine learning model is used to fit the mapping relationship, and the lower the model fitting error, the more accurate the mapping relationship fitted by the model.

[0070] Step S3: Fit a model based on the mapping relationship, filter out the confounding factors that have a significant impact on the QoS-QoE mapping relationship from the confounding factor data, and filter out the main QoS features from the QoS feature data.

[0071] In practical applications, QoE quality is affected not only by QoS but also by confounding factors. Therefore, when fitting the QoS-QoE mapping relationship, both QoS and confounding factors are used as inputs. However, due to the numerous and complex types of confounding factors, meticulously considering the influence of each type would greatly complicate the given quantity relationship analysis. Therefore, it is necessary to screen out the confounding factors that have a significant impact on the QoS-QoE mapping relationship. The impact of confounding factors can be reflected in the error of the mapping relationship model. If removing a certain type of confounding factor does not significantly affect the fitting error, it means that the confounding factor does not affect the mapping relationship and can therefore be excluded. In one embodiment, by comparing the model fitting errors with and without the target confounding factor, confounding factors that have a significant impact on the QoS-QoE mapping relationship are screened from the confounding factor data.

[0072] Specifically, using the QoS feature data and all confounding factor data as input, and QoE data as labels, a first mapping model is trained based on the machine learning model. After removing the target confounding factor, using the QoS feature data and the remaining confounding factor data as input, and QoE data as labels, a second mapping model is trained based on the machine learning model. The fitting errors of the first and second mapping models are compared. If the fitting error of the second mapping model is significantly greater than the fitting error of the first mapping model, then the target confounding factor is determined to be the significantly influential confounding factor.

[0073] As shown in the following formula:

[0074]

[0075] Equation (2)

[0076] in, This indicates the removal of target confounding factors. The second mapping model obtained by post-fitting, This represents the first mapping model obtained by fitting QoS features and all confounding factors.

[0077] If the target confounder is removed which indicates that it has a significant impact on the mapping relationship and is considered for retention. By training and testing the model with different combinations of confounders and QoS features, the influence of different confounders on the mapping relationship is analyzed, and finally several confounders with significant influence are screened out. The relationship between the input and the output is represented as:

[0078] Equation (3)

[0079] wherein, is the mapping model obtained by fitting the target confounder.

[0080] In addition, in actual applications, although different QoS features such as packet loss rate and throughput have different specific meanings, they are related and influence each other. When the network quality is poor, the quality of different QoS indicators is often poor at the same time. Since quantitative analysis needs to control variables to study the relationship between the change of a single QoS feature and QoE, the quality of the QoS feature for quantitative analysis is basically determined when the quality of other QoS features is fixed, so that when the quality of QoS features such as delay is high, the quality of throughput in the data set is almost low, and the small data variance makes it difficult to carry out quantitative analysis. To this end, in an embodiment, in order to solve the correlation problem between QoS indicators, the QoS feature analysis is performed according to the machine learning model, and the single QoS feature that has the greatest impact on QoE is selected as the QoS main feature. The QoS main feature contains the information contained in other QoS features and represents the overall network quality. By analyzing the single QoS feature that has the greatest relationship with QoE as the main feature, the mapping relationship obtained by fitting the feature approximately represents the relationship between the overall QoS feature and QoE, so the quantitative analysis can be carried out between the main QoS feature and QoE.

[0081] Specifically, the QoS main feature is selected from the QoS feature data by comparing the model fitting errors when containing a single QoS feature or all QoS features. Specifically: for each QoS feature of the QoS feature data, the single QoS feature and the confounder data are taken as the input, and the QoE data is taken as the label. A third mapping model is trained based on the machine learning model. All QoS features and confounder data are taken as input, and QoE data is taken as label. A fourth mapping model is trained based on the machine learning model. The fitting errors of each third mapping model and the fourth mapping model are compared, and the QoS feature corresponding to the third mapping model closest to the fourth mapping model in fitting error is selected as the QoS main feature.

[0082] as shown in the following formula:

[0083]

[0084] Formula (4)

[0085] By analyzing the mixed factors and the QoS characteristics, the original input is simplified as much as possible without distorting the mapping relationship information. Finally, under the influence of mixed factors, the relationship between QoS and QoE can be expressed as:

[0086] Formula (5)

[0087] Wherein, is the fourth mapping model fitted by taking all QoS characteristics and the mixed factor data as input and the QoE data as label, is a single QoS characteristic and the mixed factor data as input and the QoE data as label, and the third mapping model is fitted. is a mapping model fitted by taking the QoS main characteristics and the mixed factor data as input and the QoE data as label.

[0088] Step S4, classifying the QoS main characteristic data and the QoE data according to the mixed factors with significant influence to obtain multiple data groups, and determining the QoS-QoE quantitative relationship curve corresponding to each data group.

[0089] The influence of mixed factors is considered, and the influence is isolated by dividing the data set. In an embodiment, the QoS main characteristic data and the QoE data are classified according to the mixed factors with significant influence, and the influence effect of the mixed factors in each classification is the same, so that the influence of the mixed factors can be isolated. Referring to Figure 2 and Figure 3 , it is shown that Figure 2 shows an example of array group division; Figure 3 shows a schematic diagram of the principle of step S4.

[0090] Specifically, first, the QoS main characteristic data and the QoE data are classified according to the screened mixed factors with significant influence to obtain multiple data groups. In each data group, the influence effect of the mixed factors is the same, so formula (5) is further simplified as:

[0091] Formula (6)

[0092] Then, the QoS main feature data in each data packet is uniformly divided into multiple intervals after being sorted by value size, and the QoS and QoE quality of different data samples in each interval are approximately the same. The mean of the QoS main feature data and the mean of the QoE data in each interval are calculated to obtain multiple QoS-QoE mapping points. Based on the multiple QoS-QoE mapping points, the QoS-QoE quantitative relationship curve corresponding to the data packet is constructed. For example, each data packet is uniformly divided into 20 intervals. The QoS-QoE mapping matching represented by each interval can be obtained by calculating the mean. Finally, 20 mapping matching points can be obtained from the 20 intervals to construct a mapping relationship curve under the influence of a certain combination of mixed factors.

[0093] In addition, considering that the mapping relationship curve may have a relatively complex change trend, in order to avoid the error problem caused by curve fitting, in an embodiment, the mapping relationship curve is directly presented in a two-dimensional coordinate system, and the law of the mapping relationship curve is further studied by a clustering method.

[0094] Step S5, clustering the curves of each data packet according to the similarity of the QoS-QoE quantitative relationship curve of each data packet to obtain a change mode of the QoS-QoE quantitative relationship.

[0095] The KMeans and other clustering algorithms use the Euclidean distance to measure the similarity between curves, but cannot consider the shape difference between different curves. To this end, in order to consider the similarity between the shape and distance of the curve, in an embodiment, when clustering the curves of each data packet according to the similarity of the QoS-QoE quantitative relationship curve of each data packet, the dynamic time warping algorithm (DTW) is used to calculate the similarity between the QoS-QoE quantitative relationship curves, and the KMedoids algorithm is used to cluster the curves, and finally the change mode of the QoS-QoE quantitative relationship is obtained.

[0096] As shown in Figure 4 , the effect comparison of dynamic time warping matching and Euclidean distance matching is shown. Figure 4 It can be seen from Figure 4 that the DTW matching can consider the similarity between the distance and shape of different curves, and can cluster the quantitative relationship curves with the same change trend into one class. Through clustering, it can be concluded that there are several different change modes of the QoS-QoE quantitative relationship curve. At the same time, clustering also clusters different mixed factor groups into one class, which is convenient for observing the effect and relationship of different mixed factors.

[0097] In one embodiment, the server utilizes the changing pattern of the QoS-QoE quantitative relationship, combined with the current client's real-time QoE data and client-side profanity data, to determine matching audio and video transmission QoS parameters, thereby adjusting the audio and video data transmission between the client and the server.

[0098] The following specific example verifies the effectiveness of the method described above in this invention.

[0099] The QoS dataset used in the experiment was provided by a domestic CDN provider, while the QoE dataset was provided by one of the largest short video applications in China. The QoS-QoE mapping dataset obtained by processing the two datasets includes traffic data from 276 CDN nodes to 31 user regions across three short video domains on the three major domestic telecom operators over a period of two months, totaling approximately 4 million data points.

[0100] The average MAPE error of throughput and stuttering rate is shown in Table 1 below, comparing the existing technology based on traditional mathematical model fitting methods with the method of this invention:

[0101] Table 1

[0102]

[0103] Using the above dataset, the method of this invention can obtain the following four types of QoS-QoE relationship models: Figure 5 As shown in the figure, a) is a high-altitude type, b) is linear type, c) is bimodal type, and d) is unimodal type. It can be seen that compared with the traditional mathematical model, the method of the present invention has the lowest average MAPE error in throughput and lag rate, the best fitting effect, and the QoS-QoE quantitative relationship curve has different variation patterns.

[0104] In summary, the method for describing the QoS-QoE mapping relationship in the audio and video service disclosed in the present application combines a machine learning model to fit the QoS-QoE mapping relationship, breaks the restriction that the QoS and the QoE are not matched in level and position in the prior art, enables the server to optimize the QoS based on the real user subjective experience (QoE), and avoids the problem that the user experience may be reduced due to the traditional optimization of the QoS only. By screening the mixed factors that have a significant impact on the mapping relationship and analyzing the quantitative relationship between the QoS and the QoE according to the classification, the problem that the mapping relationship changes dynamically due to the mixed factors in the prior art is solved, the mapping relationship analysis is more accurate, and irrelevant factors are avoided. In addition, by determining the main features of the QoS, the difficulty in quantitative analysis caused by the correlation between multiple QoS features is solved, the QoS-QoE relationship is clearer, and by clustering the QoS-QoE quantitative relationship curves under different mixed factor combinations, the change mode is obtained, the influence law of the QoS on the QoE in different scenarios can be intuitively summarized, the server can match the corresponding mapping mode according to the mixed factors of the current client, dynamically adjusts the QoS optimization direction, and adapts to the mapping relationship changes in the complex network environment.

[0105] The following is a device embodiment corresponding to the above method embodiment, as shown in Figure 5 Figure 5 The structure schematic diagram of the device for describing the QoS-QoE mapping relationship in the audio and video service is shown.

[0106] The structure schematic diagram of the device for describing the QoS-QoE mapping relationship in the audio and video service is shown.

[0107] A device 300 for describing the QoS-QoE mapping relationship in the audio and video service, comprising:

[0108] The acquisition module 310 is used to collect the QoS feature data, the QoE data and the mixed factor data of the audio and video service, wherein the QoS feature data is the transmission layer performance data in the audio and video transmission process, the QoE data is the subjective experience quality data of the user to the audio and video service, and the mixed factor data is the environment related data that affects the QoS-QoE mapping relationship.

[0109] The mapping module 320 is used to combine the QoS feature data, the QoE data and the mixed factor data, adopt a machine learning model to fit the QoS-QoE mapping relationship, and obtain a mapping relationship fitting model.

[0110] ​The screening module 330 is configured to screen out the mixed factors that have a significant influence on the QoS-QoE mapping relationship from the mixed factor data and screen out the QoS main features from the QoS feature data according to the fitting model of the mapping relationship.

[0111] The classification module 340 is configured to classify the QoS main feature data and the QoE data according to the mixed factors that have a significant influence to obtain a plurality of data groups and determine a QoS-QoE quantitative relationship curve corresponding to each data group.

[0112] The clustering module 350 is configured to cluster the curves of the data groups according to the similarity of the QoS-QoE quantitative relationship curves of the data groups to obtain a change mode of the QoS-QoE quantitative relationship.

[0113] The device embodiments can be implemented in cooperation with the implementation manners of the method embodiments. The related technical details mentioned in the implementation manners of the above embodiments are still valid in the implementation manners of the method embodiments. In order to reduce repetition, they will not be described here. It should be pointed out that the scope of the method and device in the embodiments is not limited to the order of the functions performed as shown or discussed, and can also include performing the functions in a substantially simultaneous manner or in a reverse order according to the functions involved, for example, the described method can be performed in an order different from that described, and various steps can also be applied, omitted or combined. In addition, the features described with reference to some examples can be combined in other examples.

[0114] The embodiments of the present application also provide a readable storage medium, which stores a program or instructions, and the program or instructions are executed by a processor to implement the steps of the method for characterizing the QoS-QoE mapping relationship in the audio / video service, and achieve the same technical effects.

[0115] The embodiments of the present application also provide a computer program product, which includes a computer program, and the computer program is executed by a processor to implement the steps of the method for characterizing the QoS-QoE mapping relationship in the audio / video service, and achieve the same technical effects.

[0116] The processor is the processor in the electronic device in the above embodiments. The readable storage medium includes a computer readable storage medium, such as a computer readable memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, etc.

[0117] Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and the necessary general hardware platform, of course, also can be through hardware, but in many cases the former is the better embodiment. Based on such understanding, the technical solutions of the present application essentially or say the part of the prior art to make contributions can be embodied in the form of computer software program product, the computer software program product is stored in a storage medium, including a number of instructions to make a terminal (may be a mobile phone, computer, server, or network equipment, etc.) executes the method described in various embodiments of the present application.

[0118] The embodiments of the present application are described above in conjunction with the drawings, but the present application is not limited to the above-mentioned specific embodiments, the above-mentioned specific embodiments are only illustrative, but not limited, those skilled in the art can make many forms under the inspiration of the present application without departing from the purpose of the present application and the scope protected by the claims, all belong to the protection of the present application.

Claims

1. A method for characterizing QoS-QoE mapping relationships in audio and video services, characterized in that, include: Collect QoS feature data, QoE data, and confounding factor data of audio and video services. The QoS feature data is the transport layer performance data during the audio and video transmission process, the QoE data is the user's subjective experience quality data of the audio and video services, and the confounding factor data is the environmental correlation data that affects the QoS-QoE mapping relationship. By combining the QoS feature data, QoE data, and confounding factor data, a machine learning model is used to fit the QoS-QoE mapping relationship, resulting in a mapping relationship fitting model. Based on the model fitting of the mapping relationship, the confounding factors that have a significant impact on the QoS-QoE mapping relationship are screened from the confounding factor data, and the main QoS features are screened from the QoS feature data; The QoS main feature data and QoE data are classified according to the significant confounding factors to obtain multiple data groups, and the QoS-QoE quantitative relationship curve corresponding to each data group is determined. Clustering the curves of each data group based on the similarity of the QoS-QoE quantitative relationship curves of each data group yields the variation pattern of the QoS-QoE quantitative relationship.

2. The method according to claim 1, characterized in that, Combining the QoS feature data, QoE data, and confounding factor data, a machine learning model is used to fit the QoS-QoE mapping relationship, resulting in a mapping relationship fitting model, which includes: The QoS feature data, promiscuous factor data, and QoE data are sorted in chronological order. The sorted data is divided into training and test sets according to a preset ratio; Using QoS feature data and confounding factor data from the training set as input and QoE data as labels, the machine learning model is trained. The model fitting error is optimized through hyperparameter adjustment and grid search to obtain the mapping relationship fitting model. The error of the mapping relationship fitting model is verified using a test set. When the error is lower than a preset threshold, the mapping relationship fitting model is determined to be effective.

3. The method according to claim 1, characterized in that, By comparing the model fitting errors with and without the target confounding factors, confounding factors that significantly affect the QoS-QoE mapping relationship are screened from the confounding factor data, including: Using the QoS feature data and all confounding factor data as inputs, and QoE data as labels, a first mapping model is trained based on the machine learning model. After removing the target confounding factors, the second mapping model is trained based on the machine learning model, using the QoS feature data and the remaining confounding factor data as input and the QoE data as labels. By comparing the fitting errors of the first mapping model and the second mapping model, if the fitting error of the second mapping model is significantly greater than the fitting error of the first mapping model, then the target confounding factor is determined to be the confounding factor with significant influence.

4. The method according to claim 1, characterized in that, By comparing the model fitting errors when including a single QoS feature or all QoS features, the main QoS features are selected from the QoS feature data, including: For each QoS feature of the QoS feature data, a third mapping model is trained based on the machine learning model, using the single QoS feature and the confounding factor data as inputs and the QoE data as labels. Using all QoS features and the aforementioned confounding factor data as input, and QoE data as labels, a fourth mapping model is trained based on the aforementioned machine learning model. By comparing the fitting errors of each of the third mapping models with the fourth mapping model, the QoS feature corresponding to the third mapping model whose fitting error is closest to that of the fourth mapping model is selected as the QoS main feature.

5. The method according to claim 1, characterized in that, Based on the significant confounding factors, the QoS main feature data and QoE data are classified to obtain multiple data groups. A quantitative QoS-QoE relationship curve is determined for each data group, including: The QoS main feature data and QoE data are classified according to the selected confounding factors that have a significant impact, resulting in multiple data groups; In each data group, the QoS main feature data is sorted by numerical value and then evenly divided into multiple intervals; Calculate the mean of the QoS main feature data and the mean of the QoE data within each interval to obtain multiple QoS-QoE mapping points; Based on the multiple QoS-QoE mapping points, the quantitative relationship curve of QoS-QoE corresponding to the data packet is constructed.

6. The method according to claim 1, characterized in that, The curves of each data group are clustered based on the similarity of the quantitative QoS-QoE relationship curves of each data group. Specifically, the dynamic time warping algorithm is used to calculate the similarity between each quantitative QoS-QoE relationship curve, and then the KMedoids algorithm is used to cluster each curve.

7. The method according to claim 1, characterized in that, The client collects the QoE data of the audio and video service and the client-side promiscuous data, including the client's geographic location information data and access method information data; The QoS characteristic data and server-side promiscuous factor data of the audio and video service are collected on the server side. The server-side promiscuous factor data includes audio and video transmission time information data, server geographical location information data, and access operator information data. The promiscuous factor data includes client-side promiscuous factor data and server-side promiscuous factor data.

8. The method according to claim 7, characterized in that, Also includes: The server utilizes the changing pattern of the QoS-QoE quantitative relationship, combined with the real-time QoE data of the current client and the client-side miscellaneous factor data, to determine the matching audio and video transmission QoS parameters, so as to adjust the audio and video data transmission between the client and the server.

9. An apparatus for characterizing QoS-QoE mapping relationships in audio and video services, characterized in that, include: The acquisition module is used to collect QoS feature data, QoE data and confounding factor data of audio and video services. The QoS feature data is the transport layer performance data during the audio and video transmission process, the QoE data is the user's subjective experience quality data of the audio and video service, and the confounding factor data is the environmental correlation data that affects the QoS-QoE mapping relationship. The mapping module is used to combine the QoS feature data, QoE data and confounding factor data, and use a machine learning model to fit the QoS-QoE mapping relationship to obtain the mapping relationship fitting model; The filtering module is used to fit a model according to the mapping relationship, filter out the confounding factors that have a significant impact on the QoS-QoE mapping relationship from the confounding factor data, and filter out the main QoS features from the QoS feature data; The classification module is used to classify the QoS main feature data and QoE data according to the confounding factors that have a significant impact, to obtain multiple data groups, and to determine the QoS-QoE quantitative relationship curve corresponding to each data group; The clustering module is used to cluster the curves of each data group based on the similarity of the quantitative QoS-QoE relationship curves of each data group, so as to obtain the change pattern of the quantitative QoS-QoE relationship.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method of any one of claims 1-8.

Images