CN115103029B

CN115103029B - Low-rank tensor completion QoS prediction method and device based on truncated nuclear norm

Info

Publication number: CN115103029B
Application number: CN202210552295.8A
Authority: CN
Inventors: 夏虹; 陈彦萍; 董庆义; 高聪; 金小敏; 王忠民
Original assignee: Xian University of Posts and Telecommunications
Current assignee: Xian University of Posts and Telecommunications
Priority date: 2022-05-20
Filing date: 2022-05-20
Publication date: 2023-06-30
Anticipated expiration: 2042-05-20
Also published as: CN115103029A

Abstract

The invention discloses a low-rank tensor completion QoS prediction method based on a truncated kernel norm, which comprises the following steps: acquiring service quality QoS data of a service based on the service called by a user from a mobile edge service platform; preprocessing QoS data to generate sparse third-order QoS data tensors; establishing a low-rank tensor completion TLTC model based on a truncated nuclear norm by using a third-order QoS data tensor, and predicting service-missing QoS data by using the TLTC model and an alternate direction multiplier method to obtain a complete QoS data tensor; and determining the recommended service for the user according to the complete QoS data tensor. The invention predicts the missing QoS data by using the TLTC model, can better capture the time correlation between different users and different services by regularization of the cutoff kernel norms, introduces rate parameters to control the cutoff degree of all tensor modes, and finally solves the optimal solution of each variable based on a multiplier alternating direction method, thereby obtaining good estimation of the target tensor to predict the missing QoS data and ensuring the prediction precision.

Description

Low-rank tensor completion QoS prediction method and device based on truncated nuclear norm

Technical Field

The invention belongs to the technical field of information, and particularly relates to a low-rank tensor completion QoS prediction method and device based on a truncated nuclear norm.

Background

In recent years, the rapid development of mobile edge computing has led to a dramatic increase in the number of services with similar functionality. In the scenario of making service calls in a mobile edge computing environment, an edge server needs to provide relevant mobile services according to service requests issued by users. Therefore, how to select and recommend high quality services to users from a large number of candidates is a challenging problem.

QoS (Quality of Service ) is an important indicator that distinguishes between functionally equivalent quality of service levels as a non-functional attribute. Because of the large number of candidate mobile services, users have typically invoked a limited number of services and acquired corresponding QoS values in the past, but because of the lack of QoS values, no accurate recommendation can be made to the user, and predicting the lack of values based on historical QoS records is an important step in mobile service related operations.

In the related art, a collaborative filtering method is mainly adopted for predicting the missing QoS value, and mainly comprises a collaborative filtering method based on the field and a collaborative filtering method based on a model. However, in the collaborative filtering method based on the field, the user-service QoS data matrix is extremely sparse, and it is difficult to find similar users or services according to the existing QoS data, so that the accuracy of detecting similar neighbors is reduced, and the final prediction effect is affected; and, if the user has not invoked any service or the service has not been invoked by any user, the neighbor is not able to determine, and the missing value cannot be predicted using the method. However, for the collaborative filtering method based on the model, the influence of time factors on QoS data is not considered, only the relation between users and services is considered, and the method often needs to manually set some parameters, which has great influence on the final prediction accuracy.

Disclosure of Invention

In order to solve the problems in the prior art, the invention provides a low-rank tensor completion QoS prediction method and device based on a truncated kernel norm. The technical problems to be solved by the invention are realized by the following technical scheme:

in a first aspect, the present invention provides a low rank tensor complement QoS prediction method based on a truncated kernel norm, including:

acquiring service quality QoS data of a service based on the service called by a user from a mobile edge service platform;

preprocessing the QoS data to generate a sparse third-order QoS data tensor;

establishing a low-rank tensor completion TLTC model based on a truncated kernel norm by using the third-order QoS data tensor, and predicting the QoS data with the service deficiency by using the TLTC model and an alternate direction multiplier method to obtain a complete QoS data tensor;

and determining the service recommended for the user according to the complete QoS data tensor.

In one embodiment of the present invention, the step of obtaining QoS data of a service based on the service invoked by a user from a mobile edge service platform includes:

determining services called by a user from a mobile edge service platform in a plurality of time intervals of a preset time window;

QoS data of the service is acquired.

In one embodiment of the invention, the QoS data for the service includes a plurality of quadruples;

wherein the quadruple comprises: user identification, service identification, invocation time, and QoS values, including throughput and response time.

In one embodiment of the invention, the TLTC model is:

s.t.:

in the method, in the process of the invention,

representing auxiliary tensor->

Respectively represent the expansion of tensors in different k modes, k represents the order, k=1, 2,3, α _k Representing the weight cutoff parameter alpha _k ≥0，/>

Representing a functional relationship->

A portion of the observed tensor is represented,

representing the truncated nuclear norms.

In a second aspect, the present invention provides a low rank tensor completion QoS prediction apparatus based on a truncated kernel norm, comprising:

the data acquisition module is used for acquiring service quality QoS data of the service based on the service called by the user from the mobile edge service platform;

the tensor generation module is used for preprocessing the QoS data and generating a sparse third-order QoS data tensor;

the prediction module is used for establishing a low-rank tensor completion TLTC model based on a truncated nuclear norm by utilizing the third-order QoS data tensor, and predicting the QoS data with the service missing by utilizing the TLTC model and an alternate direction multiplier method to obtain a complete QoS data tensor;

and the determining module is used for determining the service recommended for the user according to the complete QoS data tensor.

Compared with the prior art, the invention has the beneficial effects that:

1. in the low-rank tensor completion QoS prediction method and device based on the truncated kernel norm, aiming at the action of time information, a third-order tensor is formed by adding a time dimension on the basis of second-order user-service so as to express QoS values, and the third-order QoS data tensor can effectively express the complex ternary relation of Qos data.

2. The invention utilizes TLTC model to predict missing QoS data, can better capture the time correlation between different users and different services through regularization of cut-off kernel norms, and considers the hidden data correlation of QoS data.

The present invention will be described in further detail with reference to the accompanying drawings and examples.

Drawings

Fig. 1 is a flowchart of a low rank tensor complement QoS prediction method based on a truncated kernel norm according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of a service invocation scenario provided by an embodiment of the present invention;

fig. 3 is a schematic diagram of a low rank tensor complement QoS prediction method based on a truncated kernel norm according to an embodiment of the present invention;

FIG. 4 is another schematic diagram of a low rank tensor complement QoS prediction method based on truncated kernel norms according to an embodiment of the present invention;

FIG. 5 (a) is a graph showing MAE test results of response time provided by the examples of the present invention;

FIG. 5 (b) is an experimental result of one RMSE for response time provided by an embodiment of the invention;

FIG. 5 (c) is an MAE experiment result of throughput provided by an embodiment of the present invention;

FIG. 5 (d) is a rmSE experiment result for throughput provided by embodiments of the present invention;

FIG. 6 (a) is another MAE experiment result of response time provided by the example of the present invention;

FIG. 6 (b) is another RMSE experiment result of the response time provided by the embodiment of the invention;

FIG. 6 (c) is another MAE experiment result of throughput provided by an embodiment of the present invention;

FIG. 6 (d) is another RMSE experiment result for throughput provided by an embodiment of the invention;

fig. 7 is a schematic structural diagram of a low rank tensor complement QoS prediction device based on a truncated kernel norm according to an embodiment of the present invention.

Detailed Description

The present invention will be described in further detail with reference to specific examples, but embodiments of the present invention are not limited thereto.

With the rapid and efficient development of cloud computing, many services are being transferred to the cloud for storage and application. MEC (Mobile edge computing ) supplements cloud computing, and an edge server is used as a service carrier of a mobile user, so that mobile service can be provided for the mobile user close to the edge computing server, and the defects of insufficient storage capacity, high time delay and low bandwidth of the user in the traditional application service are overcome. At present, a recommendation technology based on service quality prediction is widely applied to cloud computing service, so that the problem of information overload is effectively solved, but less research is still performed on application in the MEC field.

In a real cloud environment, the time when a user invokes a service is uncertain, so only a limited service invoked in the past time can be obtained, which results in sparseness in QoS data. However, it is difficult to make a correct decision when providing high quality service to users using sparse QoS data, so predicting missing QoS data has become a key step in providing high quality service to users.

In view of the above, the present invention provides a low rank tensor completion QoS prediction method and apparatus based on truncated kernel norms.

Fig. 1 is a flowchart of a low rank tensor complement QoS prediction method based on a truncated kernel norm according to an embodiment of the present invention. As shown in fig. 1, an embodiment of the present invention provides a low rank tensor complement QoS prediction method based on a truncated kernel norm, including:

s1, obtaining service quality QoS data of a service based on the service called by a user from a mobile edge service platform;

s2, preprocessing QoS data to generate sparse third-order QoS data tensors;

s3, establishing a low-rank tensor completion TLTC model based on a truncated kernel norm by using the third-order QoS data tensor, and predicting the QoS data with the service deficiency by using the TLTC model and an alternate direction multiplier method to obtain a complete QoS data tensor;

s4, determining the service recommended for the user according to the complete QoS data tensor.

It should be appreciated that the number of services invoked by a user over a period of time in the past is limited, and thus there is a lack of Qos values for the invoked services. Fig. 2 is a schematic diagram of a service invocation scenario provided in an embodiment of the present invention. Specifically, in the service invocation scenario shown in fig. 2, three users George, durant and harrden can invoke the car rental services Server1, server2 and Server3 provided by the cloud Server at the time T1 and T2; taking George as an example, the user generates Qos data of the service after each service call, but since George calls the service of Server1, server2 and Server3 at time T1 and calls the service of Server1 and Server2 at time T2, the Qos data of Server3 at time T2 is missing, that is, the Qos value of the service called by the user in a certain past time period has sparsity.

Optionally, in the step S1, the step of obtaining QoS data of a service based on the service invoked by the user from the mobile edge service platform includes:

s101, determining services called by a user from a mobile edge service platform in a plurality of time intervals of a preset time window;

s102, qoS data of the service is acquired.

In general, in a mobile service invocation environment, time is accumulated for a user to invoke a service, and after each invocation of the service, the user obtains Qos data of the invoked service. In this embodiment, the QoS data includes a plurality of quadruples < UserID, serviceID, timeID, value >, each quadruple contains a plurality of attributes of the corresponding service, where UserID is a user identifier, serviceID is a service identifier, timeID is a calling time, and Value is a QoS Value; illustratively, the QoS value includes a throughput, which is defined as the average rate of successful delivery of messages per unit time, and a response time, which is the duration between the sending of a request and the receiving of a response by a serving user.

Fig. 4 is another schematic diagram of a low rank tensor complement QoS prediction method based on a truncated kernel norm according to an embodiment of the present invention. As shown in fig. 4, after QoS data is obtained, a sparse third-order QoS data tensor is further generated. Specifically, a User with User ID 0 invokes serviceID 0 services when TimeIDs 0 and 2, and User with User ID 0 invokes Ser when

TimeIDs

0, 1 and 2For example, in the case of the service with vicID 2, qos values obtained by the same service invoked by the user in a plurality of time intervals may be stored as a time series, and a third-order Qos data tensor may be constructed using these information

Where 0 represents missing QoS data.

In this embodiment, the step of constructing the third-order QoS data tensor is as follows:

(1) QoS data < user, service,1, value > of a service called by a user is obtained;

(2) According to QoS data<user,service,1,value>Constructing a matrix N of user-services ⁽¹⁾ Taking U users at the time interval 1 as matrix rows and S called services as matrix columns;

(3) According to<user,service,1,value>Filling the corresponding QoS value in matrix N ⁽¹⁾ In (a) and (b);

(4) Constructing a matrix N of k at time intervals from 1 to k ⁽¹⁾ ,N ⁽²⁾ ,…,N ^(K) ；

(5) Obtaining third-order QoS data tensors

Wherein the user-service matrix is each slice of the tensor.

In this embodiment, the TLTC (Truncated nuclear norm Low-rank Tensor Completion, low rank tensor complement based on truncated kernel norms) model is:

s.t.:

in the method, in the process of the invention,

representation aidTensor (JavaScript of a Chinese character)>

Representing a functional relationship->

A portion of the observed tensor is represented,

representing the truncated nuclear norms.

Since the TLTC model is derived based on the matrix truncated nuclear norms, the truncated nuclear norms of the matrix will be described first. Illustratively, a given matrix

And a non-negative integer r<min (m, n), then the truncated kernel norm of matrix X may be defined as the sum of min (m, n) -r minimum singular values. However, the above procedure does not take into account the truncation of the large singular values 1 to r in the matrix, as shown in the following equation:

wherein delta _i (X) represents the ith singular value of matrix X, following the ordering rule of delta ₁ ≥δ ₂ ≥δ ₃ ≥…≥δ _min{m,n} 。

The definition of the matrix cutoff core norms cannot be directly applied to tensors of high dimensions, and thus the definition of the tensor cutoff core norms can be combined as the basis of LRTC for TNN minimization, and if the rate parameters can be set appropriately, the cutoff of each tensor mode will be automatically allocated. For third order QoS data tensors

The TNN definition model is as follows:

wherein the truncation of each tensor mode is:

k.epsilon. {1,2,3, …, d }, d represents the mode of tensor expansion, the parameter Θ is a generic rate parameter for controlling the tensor +.>

Degree of truncation at $d$modulo. And for r _k Should be satisfied that, alpha ₁ ,α ₂ ,α ₃ ,…,α _d Representing tensor->

Weight parameters when the TNN (Truncated Nuclear Norm, truncated kernel norm) is calculated by spreading into a matrix.

From the definition of the tensor cut-off kernel norms, the TLTC model for QoS data can be derived:

s.t.:

in order to ensure the dependence of the variables of tensors during tensor expansion, a set of additional constraints are added, the invention introduces an auxiliary tensor

Under specific conditions +.>

k＝1,2,3。

Further, the TLTC model was ultimately transformed as follows:

s.t.:

introducing tensors

After being an intermediate variable, the observable entries are passed to the variable +.>

k＝1,2,3。

Alternatively, the present embodiment uses ADMM (Alternating Direction Method of Multipliers, alternate direction multiplier) to optimally solve for each variable.

First, an augmented lagrangian function of the TLTC model is derived:

wherein the auxiliary variable

For updating, symbols<·,·>Representing the inner product.

Thus, the ADMM iteratively converts the original tensor completion problem into the following three sub-problems:

the relationship order of these three variables is deduced as:

wherein let ρ _k =ρ, k=, 1,2,3, i.e. variable +.>

The method comprises the following steps of: />

Is a fourth order tensor and has a size of U×S×T×3.

Calculating tensors

k=1, 2,3 and tensor->

The formulas of (a) are respectively as follows:

the following can be iterated according to the formula:

wherein, fold _k Representing tensor->

K-mode expansion of>

The singular value decomposition representing the tensor expansion matrix, U is the left singular value of the tensor expansion matrix and V is the right singular value of the tensor expansion matrix.

Wherein, the liquid crystal display device comprises a liquid crystal display device,

is a fixed constraint that can ensure the conversion of the observed information at each iteration.

Based on the above analysis, in this embodiment, the step of predicting the service-missing QoS data by using the TLTC model and the alternate direction multiplier method to obtain a complete QoS data tensor includes:

(1) Initialization of

(2) Setting initial parameters p, a _k ,γ,l＝0,

(3) Execution loop While does not converge on:

ρ＝min{1.05×ρ,ρ _max }

For k＝1,2,3then:

updating:

EndFor

updating

Updating

l＝l+1

End While

(4) Obtaining a complete QoS data tensor

In order to verify the effectiveness of the low-rank tensor complement QoS prediction method based on the truncated kernel norm, the present embodiment uses the existing 7 QoS data interpolation algorithms to compare prediction performances, and mainly researches two attributes of representative response time and throughput in QoS data.

Specifically, 7 baseline methods in the prior art are:

UPCC (User-based collaborative filtering method using Pearson Correlation Coefficient) is a classical collaborative filtering method for prediction, which predicts QoS data using similarity between similar users, and UPCC calculates similarity using Pelson coefficients;

the IPCC (Item-based collaborative filtering method using Pearson Correlation Coefficient) method predicts unknown QoS data by using the similarity between similar services, and similarly calculates the similarity by using the Pelson coefficient;

UIPCC (user-item-based CF method) is a hybrid method combining UPCC and IPCC, and utilizes similarity between similar service and similar users to predict unknown QoS data, and utilizes the Pelson coefficient to calculate the similarity;

PMF (probabilistic matrix factorization) the method provides a probability matrix decomposition model, decomposes a user service quality matrix by using Gaussian assumptions, and predicts QoS missing data;

the NTF (Non-negative Tensor Factorization) method provides a Non-negative tensor decomposition model, and a Non-negative constraint is added in the CP decomposition process of the user service time tensor model;

the WSPred (web service prediction tensor factorization) method regularizes the objective function with the average of QoS during tensor decomposition.

The ClUS (A Web Service Reliability Prediction) method adopts a k-means clustering algorithm to aggregate QoS data of the calling service in the past.

In this embodiment, a recognized large-scale data set WS-stream is selected, and a third-order QoS data tensor "user-service-time" is constructed after preprocessing an original data set, where the data set records data of two indexes of throughput and response time when 142 users call 4532 services in 64 slots (5 minutes is an interval). Table 1 shows detailed statistics of Qos attribute data, the average of throughput is 9.609kbps and the average of response time is 3.165s.

TABLE 1

Next, the QoS data set is randomly deleted to ensure sparsity of the QoS data, as shown in table 2:

TABLE 2

The present embodiment sets the density of the training data set between 10% and 30%, increments 5% each time, and defines the tensor density as the density of the training data set. For example, the training data density value is 10%, which means that 10% is randomly selected from the dataset as training data, the remaining 90% is used as test data, and 90% of the test items are predicted using 10% of the training items. In addition, the same initial hypothesis was given to all models described above, and the predicted and original values were error compared under the same training and test data set using two indices, MAE (Mean Absolute Error ) and RSME (Root Mean Squared Error, error in root mean square error).

In addition, the weight parameter α in the TLTC model _k The truncated rate parameter gamma and the learning rate ρ in the ADMM need to be initialized, and these three parameters directly affect the performance of the whole model prediction index. Alternatively, the setting of the cutoff rate parameter γ in the present embodiment is obtained by cross-validation, but the data is relatively complex and bulky, and validation is performed in the case of each missing rate, so the present embodiment performs local analysis while cross-validation of the selection parameters is performed. In particular, locallyThe analysis was based on a priori experience with a small range of parameter selections, with a cut-off rate parameter gamma of 0.01 for both throughput and response time selections. For the weight parameter alpha _k A large amount of computation is required if cross-validation is used, so alpha is set using the HaLRTC (High-accuracy Low-Rank Tensor Completion) method _k 。

The learning rate ρ in ADMM determines the convergence of the whole model, in general, larger values slow down the convergence process, while smaller values allow the model to meet the convergence in several iterations. Setting ρ=1×10 in two QoS attribute data sets of throughput and response time, respectively ^-5 Sum ρ=1×10 ^-4 The maximum number of iterations is set to 20 to achieve convergence.

The embodiment further researches the influence of the cut-off rate parameter on the prediction precision, adjusts the density of the training data set from 10% to 30%, and increases the density by 5% each time to obtain the prediction precision of the two QoS data of response time and throughput under different tensor densities. Fig. 5 (a) and 5 (b) are respectively MAE and RMSE experimental results of response time provided by the embodiment of the present invention, and fig. 5 (c) and 5 (d) are respectively MAE and RMSE experimental results of throughput provided by the embodiment of the present invention. As shown in fig. 5 (a) - (b), for the QoS data sets of throughput, the cut-off rate parameters 0.01, 0.1, 0.15, 0.2, and 0.25 were used to test, and the number of iterations was selected 20 times (learning rate), and it is apparent that the MAE value was smoother than the other drops when the integer cut-off value of the tensor was 0.01 and the tensor density was between 10% and 30%, indicating that the appearance of the difference value was better, and the MAE value was lowest, indicating that the prediction effect was also better when the cut-off rate parameter was 0.01. In RMSE, the value is less ideal when the cut-off rate parameter is 0.01, but the convergence rate is faster than that of other values, and the overall prediction effect is the best. On the other hand, when the cut-off rate parameter takes a value of 0.1, the RMSE value of the throughput is relatively low, but the MAE value is relatively high at this time.

Referring to fig. 5 (c) and 5 (d), the test is still performed using the cutoff rate parameters 0.01, 0.1, 0.15, 0.2, 0.25, with a number of iterations of 50 (learning rate ρ=1×10) ^-5 ) Convergence, obviously, with throughputSimilar to the Qos data set, the values of MAE and RMSE are the lowest when the value of the cut-off rate parameter is 0.01, so that the prediction effect is optimal.

Experimental results show that under different tensor densities, the TLTC model has smaller MAE value and RMSE value when the cutoff rate parameter is 0.01, and the cutoff rate parameter is locally verified according to prior experience, so that the prediction accuracy is improved, and the calculation cost is saved. While as the tensor density increases, the prediction accuracy of the TLTC model gradually increases with a cutoff rate parameter of 0.01, which means that higher prediction accuracy can be obtained by providing more QoS data.

Table 3 shows MAE values and RMSE values of 7 baseline methods in the prior art for the low rank tensor-complement QoS prediction method based on the truncated kernel norm provided in this embodiment. As can be seen from table 3, the neighborhood based CF method UPCC, IPCC, UIPCC and the matrix factorization PMF method have lower prediction accuracy relative to the WSPred, NTF, TLTC method using the tensor model because they only use the relationship of the user-service second order model and do not consider the ternary relationship and time information of the more useful users and services in the "user-service-time" model. Illustratively, the TLTC model provided by the invention remarkably improves the prediction precision of QoS data, and smaller MAE and RMSE values are obtained in response time and throughput under different QoS data deletion densities. The MAE and RMSE values for throughput are much greater than those for response time, since the throughput ranges from 0-1000kbps and the response time ranges from 0-20s only. As the data set density increases from 10% to 30%, the MAE and RMSE values of the TLTC method become smaller because the greater the data set density, the more information is provided for missing QoS value predictions. Therefore, the low-rank tensor complement QoS prediction method based on the truncated nuclear norm has better performance than other methods.

TABLE 3 Table 3

In addition, prediction accuracy is greatly correlated with the loss of QoS data tensor density. Fig. 6 (a) and 6 (b) are respectively the results of MAE and RMSE experiments of response time provided by the embodiments of the present invention, and fig. 6 (c) and 6 (d) are respectively the results of MAE and RMSE experiments of throughput provided by the embodiments of the present invention. The density of the training tensor was changed from 10% to 30% and increased in 5% steps, as can be seen in connection with fig. 6 (a), 6 (b):

(1) With the increase of training density, the low-rank tensor completion QoS prediction method based on the truncated nuclear norm provided by the invention has the advantages of enhanced performance, more QoS data and better prediction effect.

(2) The test result obtained by adopting the TLTC model is always better than 7 baseline methods, because the baseline method only utilizes the second-order static relation of the user-service model, and the ternary relation and time information of more useful users and services in the user-service-time model are not considered.

(3) The WSPred and NTF methods using tensor models also add time information, but predict using tensor decomposition, and do not pay attention to data loss caused by decomposition during decomposition, resulting in a decrease in prediction accuracy. The low-rank tensor complement QoS prediction method based on the truncated kernel norm provided by the invention utilizes the automatic expansion tensor mode of the truncated kernel norm, so that the global situation is considered, and the local data correlation is improved, thereby obtaining higher prediction precision.

Fig. 7 is a schematic structural diagram of a low rank tensor complement QoS prediction device based on a truncated kernel norm according to an embodiment of the present invention. As shown in fig. 7, the embodiment of the present invention further provides a low rank tensor complement QoS prediction apparatus based on a truncated kernel norm, including:

a data obtaining module 710, configured to obtain quality of service QoS data of a service invoked by a user from a mobile edge service platform based on the service;

a tensor generation module 720, configured to pre-process the QoS data to generate a sparse third-order QoS data tensor;

a prediction module 730, configured to establish a low-rank tensor completion TLTC model based on a truncated kernel norm by using the third-order QoS data tensor, and predict the service-missing QoS data by using the TLTC model and an alternate direction multiplier method to obtain a complete QoS data tensor;

a determining module 740, configured to determine a service recommended for the user according to the complete QoS data tensor.

According to the above embodiments, the beneficial effects of the invention are as follows:

In the description of the present invention, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature. In the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Further, one skilled in the art can engage and combine the different embodiments or examples described in this specification.

Although the present application has been described herein in connection with various embodiments, other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed application, from a review of the figures, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and the "a" or "an" does not exclude a plurality. A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

The foregoing is a further detailed description of the invention in connection with the preferred embodiments, and it is not intended that the invention be limited to the specific embodiments described. It will be apparent to those skilled in the art that several simple deductions or substitutions may be made without departing from the spirit of the invention, and these should be considered to be within the scope of the invention.

Claims

1. A low rank tensor completion QoS prediction method based on a truncated kernel norm, comprising:

preprocessing the QoS data to generate a sparse third-order QoS data tensor;

determining the service recommended for the user according to the complete QoS data tensor;

wherein, the TLTC model is:

in the method, in the process of the invention,

representing auxiliary tensor->

Representing a functional relationship->

Representing part of the observed tensor +.>

Representing the truncated nuclear norms.

2. The truncated kernel norm based low rank tensor completion QoS prediction method according to claim 1, wherein the step of obtaining QoS data of a service based on the service invoked by a user from a mobile edge service platform comprises:

QoS data of the service is acquired.

3. The truncated kernel norm-based low rank tensor complement QoS prediction method according to claim 2, wherein the QoS data of the service includes a plurality of quaternions;

4. A low rank tensor completion QoS prediction device based on a truncated kernel norm, comprising:

a determining module, configured to determine, according to the complete QoS data tensor, a service recommended for the user;

wherein, the TLTC model is:

in the method, in the process of the invention,

representing auxiliary tensor->

Representing a functional relationship->

Representing part of the observed tensor +.>

Representing the truncated nuclear norms.