CN118152078B

CN118152078B - Cloud desktop service dynamic configuration method and device and electronic equipment

Info

Publication number: CN118152078B
Application number: CN202410573407.7A
Authority: CN
Inventors: 马一骁
Original assignee: China Mobile Communications Group Co Ltd; China Mobile Suzhou Software Technology Co Ltd
Current assignee: China Mobile Communications Group Co Ltd; China Mobile Suzhou Software Technology Co Ltd
Priority date: 2024-05-10
Filing date: 2024-05-10
Publication date: 2024-09-03
Anticipated expiration: 2044-05-10
Also published as: CN118152078A

Abstract

The application discloses a cloud desktop service dynamic configuration method, a cloud desktop service dynamic configuration device and electronic equipment, relates to the technical field of data processing, and aims to solve the problem of poor service quality when a cloud desktop provides service for a user. The method comprises the following steps: under the condition that a first cloud desktop is configured to provide services for a target user, first information is acquired; the first information includes: the first calling service record and the first QoS value corresponding to the first calling service record; according to the first information, carrying out first adjustment on parameters of a preset deep learning model; carrying out reasoning and prediction by adopting a first adjusted preset deep learning model to obtain a second QoS value; and sending target configuration information to the target user according to the first QoS value and the second QoS value, wherein the target configuration information is used for updating a target cloud desktop for providing services for the target user. According to the embodiment of the application, the cloud desktop configuration can be dynamically adjusted according to the specific use requirement of the user in the use process of the cloud desktop, so that the QoS of the adjusted cloud desktop is improved.

Description

Cloud desktop service dynamic configuration method and device and electronic equipment

Technical Field

The present application relates to the field of data processing technologies, and in particular, to a cloud desktop service dynamic configuration method and apparatus, and an electronic device.

Background

With the growing market for cloud services (closed services), people will face numerous cloud services with similar functionality. However, quality of service (Quality of Service, qoS) of these functionally similar cloud services, for example: response time, throughput, etc. of the cloud service may vary greatly when facing different users, resulting in different user experiences.

In order to further improve the satisfaction degree of the user on the cloud service experience, in the related technology, service quality prediction is performed according to non-functional attributes of the user and the cloud service, so that cloud service recommendation is completed according to the predicted QoS value, for example, the cloud service with the highest QoS value is recommended to the user.

And the cloud desktop is used as a cloud service, and in the process of requesting the cloud desktop by a user, after a batch of cloud desktops with similar functions are determined according to the functional keywords requested by the user, efficient recommendation and distribution of the cloud desktops are required according to the non-functional attributes of the user and the cloud desktops. In the related art, a collaborative filtering mode is adopted, based on neighborhood similarity, the historical QoS value of a cloud desktop called by a user is collected to predict a new QoS value, the cloud desktop with higher QoS value is recommended to the user according to the predicted QoS value, the process is realized by using a large amount of sample data, the historical QoS value of the cloud desktop which can be obtained in reality has larger data sparsity, for a new user or a new cloud desktop, even a historical QoS value cannot be acquired, so that the QoS value cannot be predicted or the predicting effect of the QoS value predicting method in the related technology is poor, and further the cloud desktop recommended or distributed based on the predicted QoS value is not the cloud desktop with the optimal service quality, so that the service quality is poor when the cloud desktop provides services for the user.

Disclosure of Invention

The embodiment of the application provides a cloud desktop service dynamic configuration method, a cloud desktop service dynamic configuration device and electronic equipment, which can dynamically adjust cloud desktop configuration according to specific use requirements of users in the use process of a cloud desktop so as to improve QoS of the adjusted cloud desktop.

In a first aspect, an embodiment of the present application provides a cloud desktop service dynamic configuration method, where the method includes:

Under the condition that a first cloud desktop is configured to provide services for a target user, first information is acquired; wherein the first information includes: the method comprises the steps of a first calling service record and a first quality of service (QoS) value corresponding to the first calling service record; the first call service record comprises a call service record of the target user on a second cloud desktop in the cloud desktop set;

according to the first information, performing first adjustment on parameters of a preset deep learning model;

Carrying out reasoning prediction by adopting the first adjusted preset deep learning model, and obtaining a second QoS value output by the preset deep learning model, wherein the second QoS value is a QoS value when a third cloud desktop in the cloud desktop set provides service for the target user;

and sending target configuration information to the target user according to the first QoS value and the second QoS value, wherein the target configuration information is used for updating a target cloud desktop for providing services for the target user.

In a second aspect, an embodiment of the present application further provides a cloud desktop service dynamic configuration device, where the device includes:

the first acquisition module is used for acquiring first information under the condition that a first cloud desktop is configured to provide services for a target user; wherein the first information includes: the method comprises the steps of a first calling service record and a first quality of service (QoS) value corresponding to the first calling service record; the first call service record comprises a call service record of the target user on a second cloud desktop in the cloud desktop set;

the first adjustment module is used for carrying out first adjustment on parameters of a preset deep learning model according to the first information;

The first reasoning module is used for carrying out reasoning prediction by adopting the preset deep learning model after first adjustment, and obtaining a second QoS value output by the preset deep learning model, wherein the second QoS value is a QoS value when a third cloud desktop in the cloud desktop set provides service for the target user;

And the sending module is used for sending target configuration information to the target user according to the first QoS value and the second QoS value, wherein the target configuration information is used for updating a target cloud desktop for providing services for the target user.

In a third aspect, an embodiment of the present application further provides a terminal device, including: the cloud desktop service dynamic configuration method comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the steps in the cloud desktop service dynamic configuration method when executing the computer program.

In a fourth aspect, embodiments of the present application further provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps in a cloud desktop service dynamic configuration method as described above.

In a fifth aspect, embodiments of the present application also provide a computer program product comprising computer instructions which, when executed by a processor, implement the steps in a cloud desktop service dynamic configuration method as described above.

In the embodiment of the application, under the condition that a first cloud desktop is configured to provide services for a target user, first information is acquired; wherein the first information includes: the method comprises the steps of a first calling service record and a first quality of service (QoS) value corresponding to the first calling service record; the first call service record comprises a call service record of the target user for a second cloud desktop in the cloud desktop set; according to the first information, performing first adjustment on parameters of a preset deep learning model; carrying out reasoning prediction by adopting the first adjusted preset deep learning model, and obtaining a second QoS value output by the preset deep learning model, wherein the second QoS value is a QoS value when a third cloud desktop in the cloud desktop set provides service for the target user; and sending target configuration information to the target user according to the first QoS value and the second QoS value, wherein the target configuration information is used for updating a target cloud desktop for providing services for the target user. In this way, the user can dynamically acquire the call service condition of the user to the cloud desktop in the cloud desktop call process, and accordingly update the cloud desktop configuration, and the QoS of the updated cloud desktop is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments of the present application will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort to a person of ordinary skill in the art.

Fig. 1 is a flowchart of a cloud desktop service dynamic configuration method provided by an embodiment of the present application;

FIG. 2 is a schematic diagram of a data processing flow of an input layer of a preset deep learning model according to an embodiment of the present application;

FIG. 3 is a schematic diagram of a data processing flow of an intermediate layer and an output layer of a preset deep learning model in an embodiment of the present application;

FIG. 4 is a schematic diagram of a meta learning training process in an embodiment of the application;

FIG. 5 is a schematic diagram of a meta-learning test procedure in an embodiment of the present application;

fig. 6 is a flowchart of a cloud desktop recommendation process in the cloud desktop service dynamic configuration method provided by the embodiment of the present application;

Fig. 7 is a flowchart of a cloud desktop dynamic update process in the cloud desktop service dynamic configuration method provided by the embodiment of the present application;

Fig. 8 is a block diagram of a cloud desktop service dynamic configuration device provided by an embodiment of the present application;

fig. 9 is a block diagram of an electronic device according to an embodiment of the present application.

Detailed Description

The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

In order to make the embodiments of the present application more clear, the following description will be given to the related technical knowledge related to the embodiments of the present application:

In the related art, when a user requests a cloud service, a batch of cloud services are found according to functional keywords input by the user, then, qoS values of cloud services which are not used by the user are predicted through service quality prediction, so that QoS values of each cloud service are obtained, and finally, the cloud services are recommended to the user according to QoS value sorting.

Among them, the quality of service prediction in the related art mainly adopts collaborative filtering. Collaborative filtering can be categorized into neighborhood-based and model-based approaches. The neighborhood-based method predicts a new QoS value according to a historical QoS value mainly by calculating the similarity between users or cloud services. The model-based method is mainly used for training a machine learning model based on the training samples after data processing, so that the model can predict a new QoS value according to the neighborhood similarity of the samples.

It can be seen that collaborative filtering relies primarily on neighborhood similarity for prediction. However, because a cloud service QoS value which can be obtained in reality has a large data sparsity, the prediction effect of collaborative filtering is poor. In addition, the cloud service recommendation in the related art also has a cold start problem, namely when some new users participate in recommendation, the QoS value prediction model cannot be quickly adapted to the new users due to fewer history calling records similar to the new users, so that the QoS value prediction effect of using the cloud service for the new users is poor.

Compared with the related art, the embodiment of the application has at least the following advantages:

1. in the process of using the cloud desktop by the user, the configuration of the cloud desktop can be dynamically adjusted according to the actual use condition of the cloud desktop by the user;

2. In the service quality prediction process, a deep learning model is adopted to conduct feature mining between a user and a cloud desktop so as to obtain a nonlinear relation between different dimensional features of the user and the cloud desktop, and based on the nonlinear relation, the accuracy of a service quality prediction result can be greatly improved, and the number of samples required by service quality prediction can be reduced. In this way, for the new users with fewer service call records, the cloud service QoS value which can be obtained at the moment has larger data sparsity, deep feature mining can be carried out on the sparse data based on a deep learning model, and QoS value prediction with higher accuracy is realized based on the features.

The cloud desktop service dynamic configuration method, the cloud desktop service dynamic configuration device and the electronic equipment provided by the embodiment of the application are described in detail through specific embodiments and application scenes thereof by combining the attached drawings.

Referring to fig. 1, fig. 1 is a flowchart of a cloud desktop service dynamic configuration method provided by an embodiment of the present application, as shown in fig. 1, the method includes the following steps:

Step 101, acquiring first information under the condition that a first cloud desktop is configured to provide services for a target user; wherein the first information includes: the method comprises the steps of a first calling service record and a first quality of service (QoS) value corresponding to the first calling service record; the first call service record includes a call service record of the target user to a second cloud desktop in the set of cloud desktops.

In some implementations, the set of cloud desktops in the embodiments of the present application may be a set of cloud desktops that support the target cloud service function requested by the user.

In some implementations, the set of cloud desktops can be a set of all cloud desktops that the cloud desktop server can provide.

In some embodiments, the first call service record may be a call service record of the target user to any cloud desktop in the cloud desktop set in a history period, where the target user may call only a part of the cloud desktops in the cloud desktop set, that is, call the second cloud desktop, and not call the third cloud desktop in the cloud desktop set. At this time, the first QoS value when the second cloud desktop provides the service for the target user may be obtained based on the call service record of the second cloud desktop, and the QoS value when the third cloud desktop provides the service for the target user needs to be predicted by the subsequent steps.

In some implementations, the QoS values in embodiments of the application may include at least one of a network QoS value and an application QoS value.

Optionally, the network QoS value may be determined based on network metrics such as response time, bandwidth, delay, etc., for example, the response time is directly used as the network QoS value, or a certain calculation is performed on the network metrics such as response time, bandwidth, delay, etc., so as to obtain the network QoS value.

Alternatively, the application QoS value may be a QoS value for an application program inside the cloud desktop, such as video and audio quality, game data transmission stability, etc., which is mainly affected by a characteristic attribute generated when the user uses the application program inside the cloud desktop.

In some embodiments, in a scenario that a cloud desktop is dynamically configured to update in a process of invoking a cloud desktop by a user, the network QoS value and the application QoS value may be predicted, and the predicted network QoS value and the application QoS value, and the network QoS value and the application QoS value of an existing cloud desktop invoking service record of the user are combined to determine whether to update the cloud desktop allocated to the user, and allocate that cloud desktop to the user.

For example: and when the network QoS value or the application QoS value of the cloud desktop currently invoked by the user is lower than a certain threshold and the network QoS value or the application QoS value of another cloud desktop is predicted to be higher than the network QoS value or the application QoS value of the cloud desktop currently invoked by the user, the cloud desktop with the higher network QoS value or the higher application QoS value can be allocated to the user so as to replace the cloud desktop invoked by the user with the cloud desktop with the higher network QoS value or the higher application QoS value.

Step 102, according to the first information, performing a first adjustment on parameters of a preset deep learning model.

In some embodiments, the first adjustment may be: and performing feature extraction and service quality prediction on the first call service record based on a preset deep learning model to obtain a predicted QoS value, and then comparing the predicted QoS value with a first QoS value actually corresponding to the first call service record, wherein the purpose is to reduce the error between the predicted QoS value and the first QoS value actually corresponding to the corresponding first call service record, and adjusting parameters in the preset deep learning model. Therefore, the first adjusted preset deep learning model can be adapted to the service quality prediction scene under the current scene, for example, the matching degree of the first adjusted preset deep learning model and the use habit of the target user is improved, and the accuracy of the service quality prediction result when the cloud desktop is called for the target user can be improved based on the first adjusted preset deep learning model.

And 103, performing inference prediction by adopting the first adjusted preset deep learning model, and obtaining a second QoS value output by the preset deep learning model, wherein the second QoS value is a QoS value when a third cloud desktop in the cloud desktop set provides service for the target user.

In some embodiments, the third cloud desktop may be a cloud desktop in the set of cloud desktops that lacks a QoS value between the target user and the third cloud desktop, e.g., the service record of the third cloud desktop invoked by the target user is not included in the first invoking service record.

For example: based on the first information, the following QoS matrix may be constructed：

。

Wherein m is cloud desktop identification, n is user identification, qoS matrixThe elements in (a) are used to represent QoS values when a user invokes a cloud desktop, for example: element 0.04 in row 1 and column 1 in the above matrix represents a quality of service of 0.04 when a user identified as 1 invokes a cloud desktop identified as 1.

The above matrix has a portion of the elements being empty (NAN) which indicates that the first call service record does not include call records for the user and cloud desktop combination corresponding to the element, for example: the element in row 3, column 1 in the above matrix indicates that the first call service record does not include a call service record for the user identified as 1 to call the cloud desktop identified as 3. The embodiment of the application is used for predicting the partial QoS value by adopting a preset deep learning model so as to obtain a complete QoS matrix, and carrying out cloud desktop updating configuration based on the complete QoS matrix.

It should be noted that, in the embodiment of the present application, the network QoS value and the application QoS value need to be respectively established. When constructing the network QoS matrix of the cloud desktop set, the unique attribute information of the user is not considered, and therefore the vector value of the unique attribute is set to 0. And when constructing the QoS matrix of the internal application of the cloud desktop, the special attribute information of the user needs to be collected.

It is worth to put forward that in the embodiment of the application, the preset deep learning model is adopted to conduct reasoning and prediction, nonlinear relations between the user and the characteristics of different dimensions of the cloud desktop can be mined, and reasoning is conducted based on the mined characteristics, so that service quality prediction is achieved.

In some embodiments, the preset deep learning model includes an intermediate layer for extracting nonlinear relationships between features of different dimensions between a user and a cloud desktop.

In this way, compared with the method for predicting a new QoS value according to an existing QoS value based on neighborhood similarity in the related art, the service quality prediction method based on the preset deep learning model in this embodiment can use a smaller number of cloud desktop call records to perform service quality prediction, thereby solving the problem of poor service quality prediction effect caused by the data sparsity of the cloud service QoS value in the related art.

And 104, sending target configuration information to the target user according to the first QoS value and the second QoS value, wherein the target configuration information is used for updating a target cloud desktop for providing services for the target user.

In some embodiments, according to the first QoS value and the second QoS value, the cloud desktop allocated to the target user may be updated from a first cloud desktop to a cloud desktop with the highest QoS value in the cloud desktop set or a cloud desktop with the highest QoS value in a first cloud desktop subset, where the first cloud desktop subset is a cloud desktop subset in the cloud desktop set that meets an application function requirement of the target user.

In some embodiments, the cloud desktop allocated to the target user may be updated from the first cloud desktop to one of the cloud desktops set meeting the QoS requirements of the user or one of the first cloud desktop subsets meeting the QoS requirements of the user according to the first QoS value and the second QoS value, where the first cloud desktop subset is a cloud desktop subset of the cloud desktop set meeting the application function requirements of the target user.

For example: under the condition that a target user uses a first application program in a first cloud desktop, qoS requirements of the first application program can be known, then a first cloud desktop subset supporting the first application program is selected from the cloud desktop set, cloud desktops in the first cloud desktop subset are ordered according to the order of QoS values from large to small based on the first QoS value and the second QoS value, and finally the cloud desktops arranged in the first position are selected to be distributed to the target user.

In some embodiments, in the case that the target user invokes at least one application program with the currently used first cloud desktop, a second cloud desktop subset supporting the application programs may be selected from the cloud desktop set, the application programs are weighted based on habits of the target user using the application programs in the first cloud desktop, such as a duration of using a plurality of application program types in the first cloud desktop, generated data traffic, and the like, then, based on weight coefficients of the application programs, a weighting process is performed on a first QoS value or a second QoS value when each cloud desktop in the second cloud desktop subset provides services for the target user, and the cloud desktops supporting the application programs are ordered according to the order of the weighted QoS values from large to small, and finally, the cloud desktops arranged in the first position are selected and allocated to the target user.

In this embodiment, the cloud desktop configuration of the user may be dynamically adjusted according to the usage habit of the user, the user demand, and the like.

As an optional implementation manner, the first adjusting, according to the first information, a parameter of a preset deep learning model includes:

Determining first attribute information of the target user and second attribute information of the second cloud desktop according to the first call service record; wherein the first attribute information comprises first basic attribute information of the target user which is irrelevant to the application program and first specific attribute information of the target user which is relevant to the application program; the second attribute information comprises second basic attribute information of the second cloud desktop;

generating a first feature vector according to the first attribute information and the second attribute information;

inputting the first feature vector into the preset deep learning model, and obtaining a third QoS value output by the preset deep learning model;

And performing first adjustment on parameters of the preset deep learning model according to the error between the third QoS value and the first QoS value.

In some embodiments, the definition of the basic attribute information in the embodiments of the present application is as follows:

Suppose there is a set of users WhereinIs the total number of users. After removing redundant and useless information such as user ID, autonomous system number, etc., each user retains the same kind of basic attributes such as user IP, longitude, latitude, country, etc. After data processing of the basic attributes, a userCan correspond to a feature vectorWherein, the method comprises the steps of, wherein,Is the total number of user attributes. At this time, the first basic attribute information may include basic attribute information of the user such as the user IP, longitude, latitude, country, and the like.

Similar to the basic attributes of the users, the cloud desktop set is assumed to beWherein, the method comprises the steps of, wherein,Is the total number of cloud desktops that the set of cloud desktops contains, each cloud desktop retains the same kind of useful attributes, such as: cloud desktop IP, longitude, latitude, country, cloud desktop Provider (Provider), etc. After processing the data of the attribute, a cloud desktopCan be expressed as a feature vectorWherein, the method comprises the steps of, wherein,Is the total number of cloud desktop attributes. At this time, the second basic attribute information may include basic attribute information of the cloud desktop such as the cloud desktop IP, longitude, latitude, country, cloud desktop Provider, and the like.

In some implementations, the definition of the feature attribute information in the embodiments of the present application is as follows:

Because the cloud desktop is richer and more comprehensive than the functions provided by other cloud services, when a user invokes the cloud desktop, unique attributes other than the basic attributes common to the cloud services are generated. For example, the type of application frequently used by the user, such as office applications, multimedia applications, games, etc., as well as the duration of use of the user, the data traffic generated, may be recorded. These attributes dynamically affect the QoS value of the cloud desktop, so this proposal defines these attributes as unique attributes for cloud desktop quality of service predictions.

Specifically, when a userTo a cloud desktopWhen the call is made, the proposal determines the specific attribute information according to the application type used by the user. The weight of the application is calculated according to the use time length of the application and the data traffic. Then a userCan correspond to another feature vectorWhereinIs the number of application types that are to be used,Representing current user pair applicationsIs the weight obtained by normalization processing with other applications. Thereby, a userCan be expressed as: User(s) Cloud desktopThe generated one-time service invocation record may be expressed as a combination of the user and the serviceAnd generates a first QoS value corresponding to the service invocation record。

It is worth to put forward that the deep learning in the embodiment of the application trains the deep neural network by using a large number of training samples, and can learn nonlinear complex correlations among a plurality of samples, so that the model is not limited by simple linear operation in collaborative filtering. In addition, the deep neural network can embed sample features, and similarity measurement different from the traditional neighborhood is performed in a new feature space, so that the data sparsity problem can be effectively relieved.

For example: it is assumed that the deep learning model mainly includes three parts of an input layer, an intermediate layer, and an output layer, wherein the data processing flow of the input layer is shown in fig. 2, and the data processing flow of the intermediate layer and the output layer is shown in fig. 3.

As shown in fig. 2, the input layer performs data preprocessing on the user attribute information and the cloud desktop attribute information. The data type included in the attribute information may be classified into two types, one of which is discrete type data, for example, discrete type data of a user country, and the other of which is continuous type data, for example, longitude and latitude of a user position. For discrete data, one-Hot (One-Hot) encoding techniques may be employed for processing. The continuous data needs to be normalized. And after the data preprocessing is completed, extracting the characteristics of the data. The feature extraction module comprises an embedding layer for respectively embedding data with different attributes, for example, the attribute information can be expressed as the following formula:

。

wherein, The function is activated for the purpose of Relu,In order to embed the weight matrix,A bias term of initial value 0,Is the output embedded vector. In calling records for a serviceAfter all the user attribute information and the cloud service attribute information are embedded, a user feature vector and a cloud service feature vector are obtained, and then the user feature vector and the cloud service feature vector are subjected to splicing (concat) operation to obtain an input feature vector of a subsequent module, namely a first feature vector:。

intermediate and output layers are used for inputting feature vectors And thereby explore the nonlinear relationship of different dimensional features between the user and the cloud service. Specifically, as shown in fig. 3, the middle layer uses a multi-layer perceptron (Multilayer Perceptron, MLP) to operate, assuming a total of n hidden layer matrices in the MLPBias and method of making sameThen an output vector can be obtained：

;

Wherein, Representing the output vector of the intermediate layer.

The output layer is the process of further processing the intermediate output vector to obtain the final prediction result. In particular by a matrix of fully connected layersOutput vector to middle layerPerforming dimension reduction to obtain a valueThe value isThe QoS value predicted by the preset deep learning model is obtained.

In some embodiments, based on generating a first feature vector according to the first attribute information and the second attribute information, the feature vector corresponding to the first attribute information may be generatedFeature vector corresponding to the second attribute informationSplicing to obtain the feature vector of the combination of the user and the serviceI.e. the first eigenvector is. At this time, the liquid crystal display device,For representing the ith userInvoking the jth cloud desktopFeature vector at that time.

In some embodiments, the first adjustment of the parameter of the preset deep learning model according to the error between the third QoS value and the first QoS value may be to construct a loss function between the third QoS value and the first QoS value, and perform gradient descent iterative update on the parameter of the preset deep learning model based on the loss function.

For example: assuming that the preset depth model comprises an input layer, a middle layer and an output layer, wherein the input layer is used for inputting a first feature vector of a user and cloud desktop combination, the middle layer is used for mining nonlinear relations between features of different dimensions of the user and the cloud desktop according to the first feature vector, and the output layer is used for outputting the output vector of the middle layerPerforming dimension reduction processing to obtain a predicted third QoS valueAt this time, the first and second electrodes are connected,Can be expressed as the following formula:

;

wherein, Representing a full connected layer matrix in an output layer for outputting vectors to an intermediate layerPerforming dimension reduction treatment;

Then based on A first QoS value corresponding to a first feature vector input to the input layerConstructing MSE loss: ; according to the above process, the loss function of the preset deep learning model can be written as The parameters of the preset deep learning model can be calculated according to the following formula by using the loss functionPerforming gradient descent iteration update:

；

wherein, For the number of current iterations,In order for the rate of learning to be high,Representation of parametersGradient calculations were performed.

In this embodiment, based on the fact that the cloud desktop can provide richer and more comprehensive functions than other cloud services, unique attributes other than general basic attributes of the cloud services are designed, and the unique attributes can reflect user preferences of different types of application programs. By inputting the basic attribute and the special attribute of the user and the basic attribute of the cloud desktop into the preset deep learning model, the QoS value predicted by the preset deep learning model can be associated with the type of the application program used by the user for calling the cloud desktop, so that the preference of the user for the application program can dynamically image the predicted QoS value. In this way, the cloud desktop recommended or dynamically configured based on the predicted QoS value can be more matched with the user's preference for the application, and if the user attaches more importance to the application a, the cloud desktop recommended or dynamically configured based on the predicted QoS value is more biased to the cloud desktop supporting the application a.

As an alternative embodiment, the method further comprises:

under the condition that first input information from the target user is received, determining a target cloud service function requested by the target user according to the first input information;

determining a cloud desktop which supports the target cloud service function and provides services for the target user in the cloud desktop set as a fourth cloud desktop, wherein the QoS value of the cloud desktop is greater than or equal to a first threshold;

and recommending the fourth cloud desktop to the target user.

In some embodiments, the first input information may be search information input by the user in the process of searching the cloud desktop to be used, such as keywords of a target cloud service function required by the user.

In some embodiments, the first threshold may be an absolute value, e.g., the first threshold is a response time less than or equal to 1s.

In some embodiments, the first threshold may be a relative value, for example, the QoS value of the fourth cloud desktop is greater than or equal to the first threshold, and one of all cloud desktops supporting the target cloud service function with the highest QoS value may be used as the fourth cloud desktop.

In this embodiment, before the user uses the cloud desktop, the cloud desktop supporting the target cloud service function requested by the user may be recommended to the user, and the QoS value of the recommended cloud desktop is greater than or equal to the first threshold.

It should be noted that, the QoS value for providing the service for the target user may be a QoS value determined based on a service invocation record, or a QoS value predicted by using a preset deep learning model.

In some embodiments, the determining a cloud desktop of the set of cloud desktops that supports the target cloud service function and that provides services for the target user as a fourth cloud desktop having a QoS value greater than or equal to a first threshold includes:

acquiring a cloud desktop subset supporting the target cloud service function in the cloud desktop set;

obtaining second information, wherein the second information comprises: the second calling service record comprises calling service records of cloud desktops in the cloud desktop subset, which are acquired based on basic attribute information of the target user;

According to the second information, performing second adjustment on parameters of the preset deep learning model;

Performing inference prediction by adopting the second adjusted preset deep learning model, and obtaining a fifth QoS value output by the preset deep learning model, wherein the fifth QoS value is a QoS value when a fifth cloud desktop in the cloud desktop subset provides service for the target user, the cloud desktop subset comprises the fifth cloud desktop, and the second calling service record does not comprise a calling service record of the fifth cloud desktop;

And recommending a fourth cloud desktop to the target user according to the fourth QoS value and the fifth QoS value, wherein the fourth cloud desktop comprises a cloud desktop with the highest QoS value in the cloud desktop subset.

In some embodiments, the second call service record may be a call service record generated when the target user calls the cloud desktop for a historical period of time.

In some embodiments, if the target user is a new user of the cloud desktop service, and at this time, the new user does not have a call service record for the cloud desktop, then according to basic attribute information of the new user, other users similar to the new user may be found from other users who call the cloud desktop, and the call service record for the cloud desktop of the other users is used as a second call service record.

It should be noted that the second adjustment is similar to the first adjustment in the foregoing embodiment in principle and process, and includes the following differences: the input information used by the second adjustment may not include the user's unique attribute information, and the input information used by the first adjustment may include the user's unique attribute information.

In addition, the above process of performing inference prediction using the second adjusted preset deep learning model to obtain the fifth QoS value output by the preset deep learning model is similar to the process of performing inference prediction using the first adjusted preset deep learning model to obtain the second QoS value output by the preset deep learning model in the foregoing embodiment, and the difference includes: the fifth QoS value may be a network QoS value predicted based on basic attribute information of the user and the cloud desktop, and the second QoS value may include a network QoS value predicted based on basic attribute information of the user and the unique attribute new message, and a network QoS value predicted based on basic attribute information of the cloud desktop and an application QoS value.

The second adjustment and the specific process of predicting the fifth QoS value by using the preset deep learning model may refer to the explanation of the first adjustment and the second QoS value by using the preset deep learning model in the foregoing embodiment, and the specific process of predicting the fifth QoS value by using the second adjustment and the preset deep learning model will not be described herein.

In this embodiment, before the target user uses the cloud desktop, the service record may be called according to the history of the target user on the cloud desktop, or the history of other similar users on the cloud desktop, the fifth QoS value when the cloud desktop provides the service for the target user may be predicted by using the second adjusted preset deep learning model, and the cloud desktop with the highest QoS value and supporting the target cloud service function required by the user may be recommended to the user based on the fourth QoS value existing in the history call service record and the fifth QoS value missing in the history call service record.

As an optional implementation manner, the preset deep learning model is a model obtained by training based on a meta learning training mode, and the meta learning training mode comprises the following training processes:

Obtaining third information, wherein the third information comprises: the third call service record comprises call service records of at least one user of n users to at least one cloud desktop of m cloud desktops, wherein n and m are integers larger than 1;

Determining fourth information according to the third call service record, wherein the fourth information comprises: third basic attribute information of each of the n users; fourth basic attribute information of each cloud desktop in the m cloud desktops; second unique attribute information of each of the n users; wherein the third basic attribute information and the fourth basic attribute are attribute information which is not related to the application program, and the second specific attribute information is attribute information related to the application program;

Generating a training sample set according to the fourth information, wherein the training sample set comprises at least two training samples, and each training sample comprises characteristic information of one of the m cloud desktops called by one of the n users;

Dividing the training sample set into a first data set and a second data set;

Performing first training based on the first data set and the initialized deep learning model to obtain a first deep learning model;

and testing and updating parameters of the first deep learning model based on the second data set to obtain the preset deep learning model.

In some implementations, the third invocation service record may include invocation service records of the plurality of cloud desktops by the plurality of users. For example: and taking the call service record of the cloud desktop called by all users of the cloud desktop service in the historical time period as a third call service record, and at the moment, each user calls at least one cloud desktop.

In some embodiments, the meaning of the third basic attribute information is similar to the meaning of the first basic attribute information in the foregoing embodiments; the meaning of the fourth basic attribute information is similar to that of the second basic attribute information in the foregoing embodiment; the meaning of the second unique attribute information is similar to that of the first unique attribute information in the foregoing embodiment, except that it includes: the third basic attribute information, the fourth basic attribute information and the second specific attribute information are attribute information of n users and m cloud desktops, and the first basic attribute information, the second basic attribute information and the first specific attribute information are attribute information of at least one cloud desktop called by the target user.

In some embodiments, according to the fourth information, a training sample set may be generated, and a plurality of second feature vectors may be generated according to the fourth information, where one second feature vector is used to represent a feature when a user invokes a cloud desktop, and the second feature vector is associated with a sixth QoS value when the user invokes the cloud desktop.

In some embodiments, the training sample set may be divided into a first data set and a second data set at a ratio of 1:1. Of course, other ratios are also possible, such as: any of the 2:1, 3:2, etc. ratios divides the training sample set into a first data set and a second data set, and the size relationship of the first data set and the second data set is not particularly limited herein.

It should be noted that, after the first training is performed on the initialized deep learning model by using the first data set, the parameters in the initialized deep learning model are updated to obtain the first deep learning model, the first deep learning model is a model that performs best on the first data set. And the second data set is used for testing and updating parameters of the first deep learning model, so that the obtained preset deep learning model is a universal initialization model, and the model can be quickly adapted to a new sample through fewer training times.

In this embodiment, the meta learning mode is used to train and test the preset deep learning model, so that the generalization capability of the preset deep learning model can be improved, and when the preset deep learning model is applied to a new QoS value prediction scene, a small amount of training samples (i.e., training samples generated based on a small amount of call service records of a user on a cloud desktop) in the scene can be used to adjust the preset deep learning model (e.g., the first adjustment and the second adjustment in the embodiment of the present application), so that the adjusted preset deep learning model can be quickly adapted to the current QoS value prediction scene.

For ease of understanding, the following examples illustrate the process of training a preset deep learning model based on a meta learning training method in the embodiments of the present application:

1. meta learning training strategy

As shown in FIG. 4, a set of all service invocation records for a user is first obtainedThe set is then pressedIs randomly split into sets with two intersections as empty sets. One of the sets is called a support setAnother set is called a query set. The idea of meta learning is to adjust the initialization parameters of the model by using the support set sample, and then test the adjusted model by using the query set sample, thereby the initialization parameters of the current model are adjustedEvaluation and updating are performed. The support set and the query set generated by all service records of a user can be called a meta-taskAll known users can build a task pool. The deep learning model can be meta-trained by utilizing the task pool, and the specific steps are as follows:

First, initializing deep learning model parameters Then randomly selecting a batch of the task pools asIs a task of (a). For a task in the batchWith its support setInitializing parameters for a current modelUpdating once to obtain intermediate parametersWhileRemain unchanged:

；

wherein, Is the learning rate; intermediate parametersRepresenting model parameters after training based on the support set of the mth task in the task pool and prior to the update are not tested.

Support set pairs for utilizing all tasks within a batchAfter one-time cyclic update, the query set loss of all tasks in the batch is summed up, and model initialization parameters are obtainedPerforming one iteration update:

；

wherein, Representing the learning rate.

After a certain number of iterations, such as 400 times, the final model initialization parameters can be obtained。

2. Meta-learning test strategy

When a new user appears, a small number of service call records generated by the new user can be sorted to obtain a QoS matrix with more missing values, and then meta-learning test is carried out. As shown in FIG. 5 in particular, a test task is built for a new userService call records whose QoS values are known are used as test support setsUser-service combinations with missing QoS values as test query sets. First utilize test support setsFor initialization parametersUpdating the deep learning model to enable the model to be quickly adapted to new tasks:

;

wherein, Is the learning rate. The utilization parameters areCan be used for testing the query set by the deep learning model of (1)And the QoS value is predicted by the user-service combination in the cloud service recommendation.

It is worth to put forward that, in the service quality prediction method based on neighborhood similarity in the related art, there is a problem of cold start, that is, for a new user who does not have a cloud desktop calling service record or a user who has only a small number of cloud desktop calling service records, the QoS value prediction model cannot be adjusted to match with the basic attribute and the requirement of the user by using the calling service record, so that the QoS value prediction model in the related art has poor accuracy on the cloud desktop QoS value prediction result of the user, and further, the cloud desktop recommended based on the QoS value prediction result may not be the optimal choice.

In the embodiment of the application, the deep learning model is trained by adopting a meta learning mode, so that the problem that the service quality prediction model cannot be quickly adjusted to adapt to the current scene by using the existing calling service record because the calling service record of the cloud desktop is less in the process of recommending the cloud desktop or dynamically configuring and updating the cloud desktop by a user can be effectively solved. Specifically, the embodiment of the application can design meta-learning training and meta-learning testing strategies according to the service call records of the users and corresponding QoS values on the basis of the deep learning model so as to train and obtain a universal QoS value prediction model with more generalization capability, thus, for new users with more QoS values missing, the deep learning model can be quickly adapted and adjusted by using a small number of service call records of the new users without training the deep learning model from the beginning, thereby saving a great deal of time and calculation overhead.

The overall flow of the embodiment of the application is illustrated by taking a cloud desktop recommendation process before a user uses a certain cloud desktop and a cloud desktop dynamic configuration update process during the use of the cloud desktop as examples:

As shown in fig. 6, the cloud desktop recommendation process before the cloud desktop is used includes the following steps:

1. Basic attributes of a user, such as geographic position, country and the like, are acquired;

2. Acquiring a cloud desktop (not including an application) call record and QoS values known to a user, such as response time and the like;

3. the initialization deep learning model obtained after meta training is finely tuned by using the known cloud desktop calling record and QoS value, namely, the process of adapting the finely tuned deep learning model to a new task based on meta test;

4. based on the basic attribute information of the user and the basic attribute information of the cloud desktop, performing unknown QoS value prediction by utilizing the depth learning model after fine adjustment;

5. and sequencing QoS values of all cloud desktops, and outputting the optimal cloud desktop as a recommended cloud desktop.

In cloud computer projects, after a user determines a batch of cloud desktops with similar functions according to function keywords, the batch of cloud desktops are limited by resources or geographic positions, and therefore cloud service quality can be reduced or even service is unresponsive. Based on the cloud desktop recommendation process, after the user inputs the function keyword, the cloud desktop supporting the target service function corresponding to the function keyword and capable of providing the service with a higher QoS value for the user can be recommended to the user.

As shown in fig. 7, the cloud desktop dynamic configuration updating process of the user in the process of using the cloud desktop comprises the following steps:

1. Acquiring a unique attribute of a user, namely using a first weight of each application type;

2. acquiring a cloud desktop (including an application) call record and QoS values known by a user, such as video and audio quality, game data transmission stability and the like;

3. Performing secondary fine adjustment on the fine-adjusted deep learning model in the flow shown in fig. 6 by using the known cloud desktop calling record and the corresponding QoS value thereof and the first weight, so that the secondary fine-adjusted deep learning model is suitable for the subdivision task of the cloud desktop which is dynamically configured for the user at present;

4. Acquiring specific attribute information such as duration of a plurality of application types of the cloud desktop used by the new user currently, generated data flow and the like, and normalizing the specific attribute information into a second weight;

5. performing QoS value prediction according to the application use condition (including information such as a second weight, a user basic attribute, a cloud desktop basic attribute and the like) of a current user in the cloud desktop by using a secondary fine-tuned deep learning model;

6. And distributing a cloud desktop capable of meeting the current QoS value requirement to the user according to the predicted QoS value and the QoS value corresponding to the existing call service record.

Through the cloud desktop dynamic configuration updating process, after the cloud computer provider predicts the service quality of the cloud desktop by using the cloud desktop service quality prediction method based on meta learning provided by the application, the cloud desktop which is previously called by the user can be replaced by the optimal cloud desktop, so that the service quality change caused by the nonfunctional attribute of the user is responded, and an elastic service mode is realized.

According to the cloud desktop service dynamic configuration method provided by the embodiment of the application, the execution main body can be the cloud desktop service dynamic configuration device. In the embodiment of the application, the cloud desktop service dynamic configuration device is taken as an example to execute the cloud desktop service dynamic configuration method.

Referring to fig. 8, the embodiment of the present application further provides a cloud desktop service dynamic configuration device, as shown in fig. 8, where the cloud desktop service dynamic configuration device 800 includes:

A first obtaining module 801, configured to obtain first information when configuring a first cloud desktop to provide a service for a target user; wherein the first information includes: the method comprises the steps of a first calling service record and a first quality of service (QoS) value corresponding to the first calling service record; the first call service record comprises a call service record of the target user on a second cloud desktop in the cloud desktop set;

The first adjustment module 802 is configured to perform a first adjustment on parameters of a preset deep learning model according to the first information;

The first reasoning module 803 is configured to perform reasoning prediction by using the first adjusted preset deep learning model, and obtain a second QoS value output by the preset deep learning model, where the second QoS value is a QoS value when a third cloud desktop in the cloud desktop set provides service for the target user;

And a sending module 804, configured to send target configuration information to the target user according to the first QoS value and the second QoS value, where the target configuration information is used to update a target cloud desktop that provides services for the target user.

Optionally, the first adjustment module 802 includes:

The first determining unit is used for determining first attribute information of the target user and second attribute information of the second cloud desktop according to the first call service record; wherein the first attribute information comprises first basic attribute information of the target user which is irrelevant to the application program and first specific attribute information of the target user which is relevant to the application program; the second attribute information comprises second basic attribute information of the second cloud desktop;

a first generating unit, configured to generate a first feature vector according to the first attribute information and the second attribute information;

The first reasoning unit is used for inputting the first feature vector into the preset deep learning model and obtaining a third QoS value output by the preset deep learning model;

And the first adjusting unit is used for carrying out first adjustment on the parameters of the preset deep learning model according to the error between the third QoS value and the first QoS value.

Optionally, the cloud desktop service dynamic configuration apparatus 800 further includes:

the second determining module is used for determining a target cloud service function requested by the target user according to the first input information under the condition that the first input information from the target user is received;

a third determining module, configured to determine, as a fourth cloud desktop, a cloud desktop that supports the target cloud service function and provides a service for the target user in the cloud desktop set, where the QoS value is greater than or equal to a first threshold;

and the recommending module is used for recommending the fourth cloud desktop to the target user.

Optionally, the third determining module includes:

the first acquisition unit is used for acquiring a cloud desktop subset supporting the target cloud service function in the cloud desktop set;

a second obtaining unit, configured to obtain second information, where the second information includes: the second calling service record comprises calling service records of cloud desktops in the cloud desktop subset, which are acquired based on basic attribute information of the target user;

The second adjusting unit is used for carrying out second adjustment on the parameters of the preset deep learning model according to the second information;

the second reasoning unit is configured to perform reasoning prediction by using the second adjusted preset deep learning model, and obtain a fifth QoS value output by the preset deep learning model, where the fifth QoS value is a QoS value when a fifth cloud desktop in the cloud desktop subset provides services for the target user, the cloud desktop subset includes the fifth cloud desktop, and the second call service record does not include a call service record of the fifth cloud desktop;

and the recommending unit is used for recommending a fourth cloud desktop to the target user according to the fourth QoS value and the fifth QoS value, wherein the fourth cloud desktop comprises a cloud desktop with the highest QoS value in the cloud desktop subset.

Optionally, the preset deep learning model is a model trained based on a meta learning training mode, and the meta learning training mode includes the following training processes:

Dividing the training sample set into a first data set and a second data set;

Optionally, the preset deep learning model includes an intermediate layer, where the intermediate layer is used to extract a nonlinear relationship between features of different dimensions between a user and a cloud desktop.

The cloud desktop service dynamic configuration device 800 provided in the embodiment of the present application can implement each process in the foregoing cloud desktop service dynamic configuration method embodiment, and achieve the same technical effects, so that repetition is avoided, and no further description is provided herein.

The embodiment of the application also provides electronic equipment. Referring to fig. 9, fig. 9 is a block diagram of an electronic device according to an embodiment of the present application. Because the principle of solving the problem of the electronic device is similar to that of the foregoing cloud desktop service dynamic configuration method embodiment of the present application, the implementation of the electronic device may refer to the implementation of the method, and the repetition is not repeated. As shown in fig. 9, the electronic device includes: processor 900 and memory 920, processor 900 is configured to read the program in memory 920, and perform the following procedures:

Wherein in fig. 9, a bus architecture may comprise any number of interconnected buses and bridges, and in particular one or more processors represented by processor 900 and various circuits of memory represented by memory 920, linked together. The bus architecture may also link together various other circuits such as peripheral devices (e.g., cameras), voltage regulators, power management circuits, etc., as are well known in the art and, therefore, will not be described further herein. The bus interface provides an interface. The processor 900 is responsible for managing the bus architecture and general processing, and the memory 920 may store data used by the processor 900 in performing operations.

Optionally, the processor 900 is further configured to read the program in the memory 920, and perform the following procedure:

and recommending the fourth cloud desktop to the target user.

Dividing the training sample set into a first data set and a second data set;

The electronic device provided by the embodiment of the application can execute the embodiment of the cloud desktop service dynamic configuration method, and the implementation principle and the technical effect are similar, and the embodiment is not repeated here.

The embodiment of the application also provides a computer readable storage medium for storing a computer program, wherein the computer program can be executed by a processor, can realize each step of the cloud desktop service dynamic configuration method embodiment, and can obtain the same beneficial effects as the cloud desktop service dynamic configuration method embodiment, and is not repeated herein.

Embodiments of the present application also provide a computer program product stored in a nonvolatile storage medium, where the computer program product is executed by at least one processor to implement the steps of the foregoing cloud desktop service dynamic configuration method embodiment, and can achieve the same beneficial effects as the foregoing cloud desktop service dynamic configuration method embodiment, so that repetition is avoided and no further description is given here.

In the several embodiments provided in the present application, it should be understood that the disclosed methods and apparatus may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

In addition, each functional unit in the embodiments of the present application may be integrated in one processing unit, or each unit may be physically included separately, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in hardware plus software functional units.

The integrated units implemented in the form of software functional units described above may be stored in a computer readable storage medium. The software functional unit is stored in a storage medium, and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform part of the steps of the transceiving method according to the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (Random Access Memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

While the foregoing is directed to the preferred embodiments of the present application, it will be appreciated by those skilled in the art that various modifications and adaptations can be made without departing from the principles of the present application, and such modifications and adaptations are intended to be comprehended within the scope of the present application.

Claims

1. The cloud desktop service dynamic configuration method is characterized by comprising the following steps of:

According to the first information, performing first adjustment on parameters of a preset deep learning model, wherein the first adjustment is used for: performing feature extraction and service quality prediction on the first call service record based on the preset deep learning model to obtain a predicted QoS value, and then comparing the predicted QoS value with the first QoS value actually corresponding to the first call service record, so as to reduce the error between the predicted QoS value and the first QoS value actually corresponding to the first call service record, and adjusting parameters in the preset deep learning model;

2. The method of claim 1, wherein the first adjusting the parameters of the predetermined deep learning model according to the first information comprises:

3. The method according to claim 1, wherein the method further comprises:

and recommending the fourth cloud desktop to the target user.

4. The method of claim 3, wherein the determining a cloud desktop of the set of cloud desktops that supports the target cloud service function and that provides service to the target user that has a QoS value greater than or equal to a first threshold as a fourth cloud desktop comprises:

5. The method according to any one of claims 1 to 4, wherein the preset deep learning model is a model trained based on a meta learning training mode comprising the following training process:

Dividing the training sample set into a first data set and a second data set;

6. The method according to any one of claims 1 to 4, wherein the preset deep learning model comprises an intermediate layer for extracting non-linear relations between features of different dimensions between a user and a cloud desktop.

7. A cloud desktop service dynamic configuration device, comprising:

The first adjustment module is configured to perform a first adjustment on parameters of a preset deep learning model according to the first information, where the first adjustment is used for: performing feature extraction and service quality prediction on the first call service record based on the preset deep learning model to obtain a predicted QoS value, and then comparing the predicted QoS value with the first QoS value actually corresponding to the first call service record, so as to reduce the error between the predicted QoS value and the first QoS value actually corresponding to the first call service record, and adjusting parameters in the preset deep learning model;

8. The apparatus of claim 7, wherein the first adjustment module comprises:

9. The apparatus of claim 7, wherein the apparatus further comprises:

10. The apparatus of claim 9, wherein the third determination module comprises:

11. The apparatus according to any one of claims 7 to 10, wherein the preset deep learning model is a model trained based on a meta learning training mode comprising the following training process:

Dividing the training sample set into a first data set and a second data set;

12. The apparatus according to any one of claims 7 to 10, wherein the preset deep learning model comprises an intermediate layer for extracting non-linear relations between features of different dimensions between a user and a cloud desktop.

13. An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor; it is characterized in that the method comprises the steps of,

The processor is configured to read a program in a memory to implement the steps in the cloud desktop service dynamic configuration method according to any one of claims 1 to 6.

14. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the steps in the cloud desktop service dynamic configuration method according to any one of claims 1 to 6.

15. A computer program product comprising computer instructions which, when executed by a processor, implement the steps in the cloud desktop service dynamic configuration method of any of claims 1 to 6.