CN109376901B - Service quality prediction method based on decentralized matrix decomposition - Google Patents

Service quality prediction method based on decentralized matrix decomposition Download PDF

Info

Publication number
CN109376901B
CN109376901B CN201811043938.6A CN201811043938A CN109376901B CN 109376901 B CN109376901 B CN 109376901B CN 201811043938 A CN201811043938 A CN 201811043938A CN 109376901 B CN109376901 B CN 109376901B
Authority
CN
China
Prior art keywords
user
service
users
representing
information
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201811043938.6A
Other languages
Chinese (zh)
Other versions
CN109376901A (en
Inventor
刘安
彭佳
李直旭
赵雷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou University
Original Assignee
Suzhou University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Suzhou University filed Critical Suzhou University
Priority to CN201811043938.6A priority Critical patent/CN109376901B/en
Publication of CN109376901A publication Critical patent/CN109376901A/en
Application granted granted Critical
Publication of CN109376901B publication Critical patent/CN109376901B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Tourism & Hospitality (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Marketing (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Business, Economics & Management (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Health & Medical Sciences (AREA)
  • Development Economics (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses a service quality prediction method based on decentralized matrix decomposition, which comprises the following steps: (1) establishing a user adjacency graph, (2) determining the range of interaction, (3) determining the information of the interaction, and (4) synthesizing and predicting a QoS value. By adopting the mode, the service quality prediction method based on decentralized matrix decomposition predicts the QoS value by adopting the decentralized matrix decomposition method, solves the problem of computing resource waste caused by centralized training, and in addition, the QoS value of each user for Web service is stored in the user, so that the individual privacy of the user is well protected, and the service quality prediction method based on decentralized matrix decomposition has wide market prospect in popularization.

Description

Service quality prediction method based on decentralized matrix decomposition
Technical Field
The invention relates to the field of network services, in particular to a service quality prediction method based on decentralized matrix decomposition.
Background
With the rapid development of the internet technology, more and more Web services emerge, and it is increasingly difficult for internet users to find the required Web services among the massive Web services. Therefore, more and more Web service recommendation systems have come into existence and have received extensive attention and research.
In order to implement personalized Web service recommendation, a Web service recommendation system needs to collect user evaluation conditions on quality of service (qos). QoS is widely used to describe non-functional features of Web services, such as: response time, throughput, price, reliability, etc. Therefore, the QoS-based service recommendation system has been widely applied in the real society. One common assumption in these studies is that the QoS value of a Web service is always present.
However, it is impractical for the system to obtain the QoS value of each Web service that the user evaluates, on one hand, the QoS value published by the Web service provider or third party community is inaccurate for the user because the actual QoS value is susceptible to uncertain network environment and the area where the user is located; on the other hand, due to limitations of other resources such as time, cost, etc. How to obtain an accurate QoS value is a major issue. In order to solve the problem, various Web service QoS prediction methods are proposed, wherein a matrix decomposition method is one of the commonly used technologies, and has high accuracy and good performance in many recommendation applications.
The existing matrix decomposition Web service QoS prediction methods are all centralized training methods, specifically, a platform of a recommendation system is constructed firstly, QoS values of all users for all Web services are collected, and then a matrix decomposition model is constructed by using the data. There are several drawbacks to doing so:
(1) the storage resources are wasted, and the recommendation system needs to collect the QoS values of all users, so that the QoS values of all users for Web service need to be stored on a certain server in a centralized manner;
(2) computing resources are wasted, when the matrix decomposition model is trained, the model needs to be trained on a server, and the training speed of the model is limited by the number of machines at a server side;
(3) the privacy data of the user cannot be protected, the QoS value of the user to the Web service is acquired by the server, the preference information of the user may be revealed to a malicious attacker, and the problem of potential privacy safety hazards of the user exists.
Disclosure of Invention
The technical problem mainly solved by the invention is to provide a service quality prediction method based on decentralized matrix decomposition, which predicts QoS values by adopting the decentralized matrix decomposition method, on one hand, data of users are stored on personal equipment of the users without being uploaded to a server, so that the problem of storage resource waste caused by a centralized training model is solved, on the other hand, the training of the model is also completed at a user side, the cooperative training of the model is completed among the users by interacting non-original data information, so that the problem of computing resource waste caused by the centralized training can be solved, and in addition, the QoS values of each user to Web services are stored in the hands of the users, so that the individual privacy of the users is well protected, and the service quality prediction method based on decentralized matrix decomposition has wide market prospect in popularization.
In order to solve the technical problem, the invention provides a service quality prediction method based on decentralized matrix decomposition, which comprises the following steps:
definition of
Figure DEST_PATH_IMAGE002
Representing a set of users, defining
Figure DEST_PATH_IMAGE004
Representing a collection of Web services in common
Figure DEST_PATH_IMAGE006
A user and
Figure DEST_PATH_IMAGE008
a Web service that is a Web service of a Web service,
Figure DEST_PATH_IMAGE010
representing the interaction information of the ith user to the jth Web service, wherein
Figure DEST_PATH_IMAGE012
And represents the user
Figure DEST_PATH_IMAGE014
To the service
Figure DEST_PATH_IMAGE016
The value of the QoS being evaluated is,
definition of
Figure DEST_PATH_IMAGE018
Representing a user latent feature matrix, wherein each row
Figure DEST_PATH_IMAGE020
Representing a user
Figure 893612DEST_PATH_IMAGE014
Is/are as follows
Figure DEST_PATH_IMAGE022
Dimension latent feature vector, definition
Figure DEST_PATH_IMAGE024
Tensor representing latent features of service, definition
Figure DEST_PATH_IMAGE026
Representing a user
Figure 466545DEST_PATH_IMAGE014
Of service latent feature matrix, wherein
Figure DEST_PATH_IMAGE028
Representing a user
Figure 636495DEST_PATH_IMAGE014
To the service
Figure 562863DEST_PATH_IMAGE016
The K-dimensional potential feature vector of (a),
(1) establishing a user adjacency graph:
building user adjacency graphs based on geographic location of users
Figure DEST_PATH_IMAGE030
To express the degree of closeness between users, define
Figure DEST_PATH_IMAGE032
Representing a user
Figure 236289DEST_PATH_IMAGE014
And the user
Figure DEST_PATH_IMAGE034
Of the user, then
Figure 324331DEST_PATH_IMAGE014
And the user
Figure 997889DEST_PATH_IMAGE034
The similarity between them can be expressed as:
Figure DEST_PATH_IMAGE036
，
wherein:
Figure DEST_PATH_IMAGE038
if the user is
Figure 711636DEST_PATH_IMAGE014
And the user
Figure 911673DEST_PATH_IMAGE034
In the same region, in
Figure DEST_PATH_IMAGE040
If the user is
Figure 295250DEST_PATH_IMAGE014
And the user
Figure 580738DEST_PATH_IMAGE034
Not in the same area, in the formula
Figure DEST_PATH_IMAGE042
，
Figure DEST_PATH_IMAGE044
Is a mapping function of the distance between users and the similarity, the smaller the distance between users, the greater the similarity,
(2) determining the range of interaction:
the user adopts a method of adjacent user interaction based on random walk to interact with the information of other neighbors:
after the user adjacency graph is established
Figure 176804DEST_PATH_IMAGE030
Thereafter, a user adjacency matrix is employed
Figure DEST_PATH_IMAGE046
Representing adjacency graphs, definitions, of users
Figure DEST_PATH_IMAGE048
In the representation
Figure 372293DEST_PATH_IMAGE014
To (1) a
Figure DEST_PATH_IMAGE050
The number of the neighbors is one,
Figure DEST_PATH_IMAGE052
which indicates the number of neighbors to be present,
Figure DEST_PATH_IMAGE054
it is clear that, in the case of a,
Figure DEST_PATH_IMAGE056
indicate the user
Figure 926772DEST_PATH_IMAGE014
The direct neighbourhood of (a) the network,
when the user is
Figure 965135DEST_PATH_IMAGE014
Want to be in direct neighborhood with him
Figure DEST_PATH_IMAGE058
The information is exchanged between the mobile terminal and the mobile terminal,
Figure DEST_PATH_IMAGE060
representing a user
Figure 161630DEST_PATH_IMAGE014
The behavior of a user is selected from his neighborhood, and then,
Figure DEST_PATH_IMAGE062
，
according to Markov probability, user
Figure 664155DEST_PATH_IMAGE014
Choose his second neighbor (
Figure DEST_PATH_IMAGE064
) The probability of (c) is:
Figure DEST_PATH_IMAGE066
，
wherein D represents the maximum distance of random walk,
the user walks around the neighbors within a preset maximum number of iterations to exchange information,
(3) determining the interactive information:
in order to protect the privacy of users, to determine which information is interacted between users, the QoS value is decomposed into potential vectors of user preference
Figure 468163DEST_PATH_IMAGE020
And service preference potential vector
Figure 852877DEST_PATH_IMAGE028
That is to say that,
Figure DEST_PATH_IMAGE068
，
wherein for each user
Figure 603795DEST_PATH_IMAGE014
Of 1 at
Figure DEST_PATH_IMAGE070
Service latent feature vector of individual service
Figure 632931DEST_PATH_IMAGE028
Can be decomposed into:
Figure DEST_PATH_IMAGE072
，
wherein
Figure DEST_PATH_IMAGE074
A global latent feature vector is represented, which represents a common preference of all users,
Figure DEST_PATH_IMAGE076
representing private latent feature vectors, which represent the personal preferences of the user, the loss function can be expressed as:
Figure DEST_PATH_IMAGE078
，
(Vector)
Figure 388266DEST_PATH_IMAGE020
and
Figure 73326DEST_PATH_IMAGE076
relying only on storage at the user
Figure 408361DEST_PATH_IMAGE014
Information in (1), and
Figure 495266DEST_PATH_IMAGE074
not only dependent on the user
Figure 968972DEST_PATH_IMAGE014
The information in (1) also depends on information in other adjacent users, and a privacy protection protocol is set to exchange information between the users, so that the global potential feature vector of the service is learned
Figure 382DEST_PATH_IMAGE074
The privacy protection protocol is to send the gradient of the loss function of each user to its neighbors
Figure DEST_PATH_IMAGE080
To learn global service latent feature vectors
Figure DEST_PATH_IMAGE082
For each user
Figure 545633DEST_PATH_IMAGE014
Loss function of
Figure DEST_PATH_IMAGE084
About
Figure 221465DEST_PATH_IMAGE020
，
Figure 866073DEST_PATH_IMAGE074
And
Figure 525724DEST_PATH_IMAGE076
the gradient of (d) is:
Figure DEST_PATH_IMAGE086
Figure DEST_PATH_IMAGE088
Figure DEST_PATH_IMAGE090
，
(4) synthesizing and predicting QoS values:
the user after the interaction is completed will make the global latent feature vector
Figure 327196DEST_PATH_IMAGE074
And private latent feature vectors
Figure 795218DEST_PATH_IMAGE076
Synthesizing to obtain service latent feature vectors
Figure 813989DEST_PATH_IMAGE028
I.e. by
Figure DEST_PATH_IMAGE072A
，
Then the user preference potential vector is used
Figure 210204DEST_PATH_IMAGE020
And service preference potential vector
Figure 644728DEST_PATH_IMAGE028
Synthesizing to obtain users
Figure 357469DEST_PATH_IMAGE014
QoS value evaluated for service, i.e.
Figure DEST_PATH_IMAGE068A
。
In a preferred embodiment of the invention, the QoS value and potential characteristics of each user for a service are stored at each user, i.e. each user
Figure 734093DEST_PATH_IMAGE014
Only his K-dimensional potential feature vector needs to be saved
Figure 40440DEST_PATH_IMAGE020
And service latent feature matrix
Figure 340971DEST_PATH_IMAGE026
。
In a preferred embodiment of the invention, the attributes of the QoS value include response time and throughput.
The invention has the beneficial effects that: the service quality prediction method based on decentralized matrix decomposition predicts the QoS value by adopting the decentralized matrix decomposition method, on one hand, the data of the user is stored on personal equipment of the user without being uploaded to a server side, so that the problem of storage resource waste caused by a centralized training model is solved, on the other hand, the training of the model is also completed at the user side, the cooperative training of the model is completed among the users by interacting non-original data information, the problem of computing resource waste caused by the centralized training can be solved, in addition, the QoS value of each user to Web service is stored in the hands of the user, so that the individual privacy of the user is well protected, and the service quality prediction method based on decentralized matrix decomposition has wide market prospect in popularization. .
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without inventive efforts, wherein:
FIG. 1 is a user-service call graph of a preferred embodiment of the decentralized matrix factorization based QoS prediction method of the present invention;
FIG. 2 is an observed QoS matrix of a preferred embodiment of the decentralized matrix factorization based QoS prediction method of the present invention;
FIG. 3 is a predicted QoS matrix for a preferred embodiment of the decentralized matrix factorization based QoS prediction method of the present invention;
FIG. 4 is a decomposition diagram of a prior art centralized matrix decomposition method;
FIG. 5 is a decomposition diagram of a preferred embodiment of the method for QoS prediction based on decentralized matrix decomposition according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1 to 5, an embodiment of the present invention includes:
a service quality prediction method based on decentralized matrix decomposition comprises the following steps:
definition of
Figure 173798DEST_PATH_IMAGE002
Representing a set of users, defining
Figure 190165DEST_PATH_IMAGE004
Representing a collection of Web services in common
Figure 374021DEST_PATH_IMAGE006
A user and
Figure 150347DEST_PATH_IMAGE008
a Web service that is a Web service of a Web service,
Figure 572102DEST_PATH_IMAGE010
representing the interaction information of the ith user to the jth Web service, wherein
Figure 103577DEST_PATH_IMAGE012
And is and
Figure DEST_PATH_IMAGE092
representing a user
Figure 899364DEST_PATH_IMAGE014
To the service
Figure 744960DEST_PATH_IMAGE016
The value of the QoS being evaluated is,
definition of
Figure 286800DEST_PATH_IMAGE018
Representing a user latent feature matrix, wherein each row
Figure 933985DEST_PATH_IMAGE020
Representing users
Figure 561276DEST_PATH_IMAGE022
Dimension latent feature vector, definition
Figure 7301DEST_PATH_IMAGE024
Tensor representing latent features of service, definition
Figure 279013DEST_PATH_IMAGE026
Representing a user
Figure 949029DEST_PATH_IMAGE014
Of service latent feature matrix, wherein
Figure 188249DEST_PATH_IMAGE028
Representing a user
Figure 402412DEST_PATH_IMAGE014
To the service
Figure 840215DEST_PATH_IMAGE016
The K-dimensional potential feature vector of (a),
(1) establishing a user adjacency graph:
building user adjacency graphs based on geographic location of users
Figure 87657DEST_PATH_IMAGE030
To express the degree of closeness between users, define
Figure 814174DEST_PATH_IMAGE032
Representing a user
Figure 241482DEST_PATH_IMAGE014
And the user
Figure 815682DEST_PATH_IMAGE034
Of the user, then
Figure 93080DEST_PATH_IMAGE014
And the user
Figure 792046DEST_PATH_IMAGE034
The similarity between them can be expressed as:
Figure DEST_PATH_IMAGE036A
，
wherein:
Figure 570515DEST_PATH_IMAGE038
if the user is
Figure 795960DEST_PATH_IMAGE014
And the user
Figure 854046DEST_PATH_IMAGE034
In the same region, in
Figure 696100DEST_PATH_IMAGE040
If the user is
Figure 747101DEST_PATH_IMAGE014
And the user
Figure 30315DEST_PATH_IMAGE034
Not in the same area, in the formula
Figure 649515DEST_PATH_IMAGE042
，
Figure 323073DEST_PATH_IMAGE044
Is a mapping function of the distance between users and the similarity, the smaller the distance between users, the greater the similarity,
(2) determining the range of interaction:
the user adopts a method of adjacent user interaction based on random walk to interact with the information of other neighbors:
after the user adjacency graph is established
Figure 318711DEST_PATH_IMAGE030
Thereafter, a user adjacency matrix is employed
Figure 377803DEST_PATH_IMAGE046
Representing adjacency graphs, definitions, of users
Figure 167904DEST_PATH_IMAGE048
To represent
Figure 328758DEST_PATH_IMAGE030
In
Figure 800191DEST_PATH_IMAGE014
To (1) a
Figure 776106DEST_PATH_IMAGE050
The number of the neighbors is one,
Figure 940371DEST_PATH_IMAGE052
which indicates the number of neighbors to be present,
Figure 447576DEST_PATH_IMAGE054
it is clear that, in the case of a,
Figure 394803DEST_PATH_IMAGE056
indicate the user
Figure 569432DEST_PATH_IMAGE014
The direct neighbourhood of (a) the network,
when the user is
Figure 560391DEST_PATH_IMAGE014
Want to be in direct neighborhood with him
Figure 86050DEST_PATH_IMAGE058
The information is exchanged between the mobile terminal and the mobile terminal,
Figure 571389DEST_PATH_IMAGE060
representing a user
Figure 600525DEST_PATH_IMAGE014
The behavior of a user is selected from his neighborhood, and then,
Figure DEST_PATH_IMAGE062A
，
according to Markov probability, user
Figure 27964DEST_PATH_IMAGE014
Choose his second neighbor (
Figure 447444DEST_PATH_IMAGE064
) The probability of (c) is:
Figure DEST_PATH_IMAGE066A
，
wherein D represents the maximum distance of random walk,
the user walks around the neighbors within a preset maximum number of iterations to exchange information,
(3) determining the interactive information:
in order to protect the privacy of users, to determine which information is interacted between users, the QoS value is decomposed into potential vectors of user preference
Figure 657846DEST_PATH_IMAGE020
And a service preference potential vector, i.e.,
Figure DEST_PATH_IMAGE068AA
，
wherein for each user
Figure 197281DEST_PATH_IMAGE014
Of 1 at
Figure 670987DEST_PATH_IMAGE070
Service latent feature vector of individual service
Figure 781026DEST_PATH_IMAGE028
Can be decomposed into:
Figure DEST_PATH_IMAGE072AA
，
where a global latent feature vector is represented, which represents a common preference of all users,
Figure 919752DEST_PATH_IMAGE076
representing private latent feature vectors representing users
Figure 798846DEST_PATH_IMAGE014
Then the loss function may be expressed as:
Figure DEST_PATH_IMAGE078A
，
(Vector)
Figure 568088DEST_PATH_IMAGE020
and rely only on being stored at the user
Figure 24477DEST_PATH_IMAGE014
Information in (1), and
Figure 655310DEST_PATH_IMAGE074
not only dependent on the user
Figure 779123DEST_PATH_IMAGE014
In the method, a privacy protection protocol is set to exchange information among users so as to learn the global latent feature vector of the service by sending information in each user and also relying on information in other adjacent users
Figure 719267DEST_PATH_IMAGE014
Loss function of
Figure 662952DEST_PATH_IMAGE084
About
Figure 831896DEST_PATH_IMAGE074
To his neighbors
Figure 747899DEST_PATH_IMAGE080
To learn global service latent feature vectors
Figure 655681DEST_PATH_IMAGE082
For each user
Figure 86663DEST_PATH_IMAGE014
Loss function of
Figure 387194DEST_PATH_IMAGE084
About
Figure 829808DEST_PATH_IMAGE020
，
Figure 987120DEST_PATH_IMAGE074
And
Figure 30031DEST_PATH_IMAGE076
the gradient of (d) is:
Figure DEST_PATH_IMAGE086A
Figure DEST_PATH_IMAGE088A
Figure DEST_PATH_IMAGE090A
，
(4) synthesizing and predicting QoS values:
the user after the interaction is completed will make the global latent feature vector
Figure 406958DEST_PATH_IMAGE074
And private latent feature vectors
Figure 94292DEST_PATH_IMAGE076
Synthesizing to obtain service latent feature vectors
Figure 297871DEST_PATH_IMAGE028
I.e. by
Figure DEST_PATH_IMAGE072AAA
，
Then the user preference potential vector is used
Figure 93658DEST_PATH_IMAGE020
And service preference potential vector
Figure 798308DEST_PATH_IMAGE028
Synthesizing to obtain users
Figure 215514DEST_PATH_IMAGE014
To the service
Figure 714629DEST_PATH_IMAGE016
Evaluated QoS value, i.e.
Figure DEST_PATH_IMAGE068AAA
。
Preferably, the QoS value and potential characteristics of each user to service are both maintained at each user, i.e. each user
Figure 669815DEST_PATH_IMAGE014
Only his K-dimensional potential feature vector needs to be saved
Figure 646999DEST_PATH_IMAGE020
And service latent feature matrix
Figure 433558DEST_PATH_IMAGE026
。
Preferably, the attributes of the QoS value include response time and throughput.
The present example is implemented on the premise of the technical solution of the present invention, and a detailed implementation manner and process are given, but the scope of the present invention is not limited to the following examples.
The invention tests on publicly available QoS data sets containing 339 user ratings 5825 QoS values for Web services, mainly considering two representative QoS attributes: response Time (RT), which represents the duration between the user making the request and receiving the response, and Throughput (TP), which represents the data transfer rate at which the user invokes the service. The details of the data set are as follows;
QoS user' s Service Range Mean value of Variance (variance) Density of data
RT(sec) 339 5825 0~20 0.909 1.973 94.8%
TP(kbps) 339 5825 0~1000 47.562 110.797 92.7%
In the experiment, the QoS value of the user calling service is 339
Figure DEST_PATH_IMAGE094
5825, where-1 represents unknown QoS values, i.e., QoS values that need to be predicted, this experiment randomly deletes some values, and only maintains QoS values for small data densities, such as: the data density =10% indicates that the user only evaluates the QoS value for 10% of the Web services, predicts the remaining QoS value by the QoS prediction algorithm of the decentralized matrix factorization of the present invention, and evaluates the accuracy of QoS prediction by RMSE (root mean square error):
Figure DEST_PATH_IMAGE096
wherein the content of the first and second substances,
Figure DEST_PATH_IMAGE098
representing observed users
Figure DEST_PATH_IMAGE100
To the service
Figure DEST_PATH_IMAGE102
The value of the QoS being evaluated is,
Figure DEST_PATH_IMAGE104
representing predicted users
Figure DEST_PATH_IMAGE106
To the service
Figure DEST_PATH_IMAGE108
The QoS value of (a) of (b),
Figure DEST_PATH_IMAGE110
the number of all QoS values to be predicted in the training set is represented, and the smaller the RMSE is, the higher the prediction accuracy is.
To verify the feasibility and effectiveness of the present invention, this experiment compares the inventive QoS prediction method based on Decentralized Matrix Factorization (DMF) with the following three QoS prediction methods:
MF: the method is the most basic QoS and side method based on matrix decomposition;
RMF: the QoS prediction method is based on data random fuzzy privacy protection, users randomly fuzz their QoS values by using a data randomization technology, specifically, each user adds a random number in a certain range to an original QoS value, and then sends the QoS prediction to a recommendation system set for QoS prediction.
LMF: the QoS prediction method for protecting privacy is based on a differential privacy technology.
The invention is found to perform well in predicting the QoS value of the user to the Web service which is not evaluated. In the execution process, the data of the user is stored at the user side and does not need to be uploaded to a service recommendation system, so that the waste of storage resources is reduced; the training of the model is also completed at the user side, and the cooperative training of the model is completed among users through the interaction gradient, so that the waste of computing resources is reduced; the QoS value of each user for the Web service is stored at the user side, so that the personal privacy of the user is well protected; and the method also has good effect on the aspect of prediction accuracy. The comparative effect of the method in 4 above is as follows;
Method RMSE（K=5） RMSE（K=10） RMSE（K=15） RMSE（K=20）
MF 219.59 193.62 162.39 143.83
RMF 415.90 365.71 305.54 272.11
LMF 277.65 240.81 206.16 180.85
DMF 108.80 106.48 105.97 103.20
the service quality prediction method based on decentralized matrix decomposition has the advantages that:
the QoS value is predicted by adopting a decentralized matrix decomposition method, on one hand, data of users are stored on personal equipment of the users without being uploaded to a server side, so that the problem of storage resource waste caused by a centralized training model is solved, on the other hand, training of the model is also completed at a user side, and cooperative training of the model is completed among the users by interacting non-original data information, so that the problem of computing resource waste caused by the centralized training can be solved, and in addition, the QoS value of each user to the Web service is stored in the hands of the users, so that the personal privacy of the users is well protected.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by the present specification, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (3)

1. A service quality prediction method based on decentralized matrix decomposition is characterized by comprising the following steps:
definition of
Figure FDA0003064386240000011
Representing a set of users, defining
Figure FDA0003064386240000012
Representing a collection of Web services, with a total of U users and S Web services, (U)i,sj) Representing the interaction information of the ith user to the jth Web service, wherein
Figure FDA0003064386240000013
And r isijRepresenting user uiTo service sjThe value of the QoS being evaluated is,
definition PU×KRepresenting a user latent feature matrix, wherein each row piRepresenting potential vectors of user preferences, defining QU×K×STensor representing latent features of service, definition
Figure FDA0003064386240000014
Representing user uiOf service latent feature matrix, wherein
Figure FDA0003064386240000015
A potential vector representing the service preferences,
(1) establishing a user adjacency graph:
building user adjacency graphs based on geographic location of users
Figure FDA0003064386240000016
To express the degree of closeness between users, d is definedi,i'Representing user uiAnd user ui‘Of the user u, theniAnd user ui‘The similarity between them can be expressed as:
ωi,i'＝Ii,i'f(di,i')，
wherein: omegai,i'∈[0,1]If user uiAnd user ui‘In the same region, formula Ii,i'1 if user uiAnd user ui‘Not in the same area, in the formula Ii,i'＝0，f(di,i') Is a mapping function of the distance between users and the similarity, the smaller the distance between users, the greater the similarity,
(2) determining the range of interaction:
the user adopts a method of adjacent user interaction based on random walk to interact with the information of other neighbors:
after the user adjacency graph is established
Figure FDA0003064386240000017
Then, the user adjacency graph of the user is represented by the user adjacency matrix W, and definition is carried out
Figure FDA0003064386240000021
To represent
Figure FDA0003064386240000022
Middle uiThe d-th neighbor of (a) is,
Figure FDA0003064386240000023
which indicates the number of neighbors to be present,
Figure FDA0003064386240000024
it is clear that,
Figure FDA0003064386240000025
represent user uiThe direct neighbourhood of (a) the network,
when user uiWant to be in direct neighborhood with him
Figure FDA0003064386240000026
Interaction information, XiRepresenting user uiThe behavior of a user is selected from his neighborhood, and then,
Figure FDA0003064386240000027
user u according to Markov probabilityiChoose his second neighbor
Figure FDA0003064386240000028
The probability of (c) is: p (X)i＝uk')＝∑kP(Xi＝uk)P(Xi＝uk')∝∑kωi,kωk,k'，
Wherein D represents the maximum distance of random walk,
the user walks around the neighbors within a preset maximum number of iterations to exchange information,
(3) determining the interactive information:
in order to protect the privacy of users, to determine which information is interacted between users, the QoS value is decomposed into potential vectors p of user preferenceiAnd service preference potential vector
Figure FDA0003064386240000029
That is to say that the first and second electrodes,
Figure FDA00030643862400000210
wherein for each user uiService preference potential vector for jth service
Figure FDA00030643862400000211
Can be decomposed into:
Figure FDA00030643862400000212
wherein
Figure FDA00030643862400000213
A global latent feature vector is represented, which represents a common preference of all users,
Figure FDA00030643862400000214
represent private latent feature vectors, which represent user uiThen the loss function may be expressed as:
Figure FDA00030643862400000215
vector piAnd
Figure FDA00030643862400000216
relying only on storage in user uiInformation in (1), and
Figure FDA00030643862400000217
not only dependent on user uiThe information in (1) also depends on information in other adjacent users, and a privacy protection protocol is set to exchange information between the users, so that the global potential feature vector of the service is learned
Figure FDA0003064386240000031
The privacy protection protocol is implemented by sending each user uiLoss function of
Figure FDA0003064386240000032
About
Figure FDA0003064386240000033
To his neighbors ui'To learn global service preference potential vectors
Figure FDA0003064386240000034
For each user uiLoss function of
Figure FDA0003064386240000035
With respect to pi，
Figure FDA0003064386240000036
And
Figure FDA0003064386240000037
the gradient of (d) is:
Figure FDA0003064386240000038
Figure FDA0003064386240000039
Figure FDA00030643862400000310
(4) predicting the QoS value:
the user after the interaction is completed will make the global latent feature vector
Figure FDA00030643862400000311
And private latent feature vectors
Figure FDA00030643862400000312
Synthesizing to obtain service preference potential vectors
Figure FDA00030643862400000313
Namely, it is
Figure FDA00030643862400000314
Then the user preference potential vector piAnd service preference potential vector
Figure FDA00030643862400000315
Synthesizing to obtain user uiTo service sjEvaluated QoS value, i.e.
Figure FDA00030643862400000316
2. The method of claim 1, wherein the QoS value and potential characteristics of each user for the service are stored at each user, i.e. each user uiOnly his user preference potential vector p needs to be savediAnd service latent feature matrix
Figure FDA00030643862400000317
3. The method of claim 1, wherein the attributes of the QoS value include response time and throughput.
CN201811043938.6A 2018-09-07 2018-09-07 Service quality prediction method based on decentralized matrix decomposition Active CN109376901B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811043938.6A CN109376901B (en) 2018-09-07 2018-09-07 Service quality prediction method based on decentralized matrix decomposition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811043938.6A CN109376901B (en) 2018-09-07 2018-09-07 Service quality prediction method based on decentralized matrix decomposition

Publications (2)

Publication Number Publication Date
CN109376901A CN109376901A (en) 2019-02-22
CN109376901B true CN109376901B (en) 2021-09-03

Family

ID=65404858

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811043938.6A Active CN109376901B (en) 2018-09-07 2018-09-07 Service quality prediction method based on decentralized matrix decomposition

Country Status (1)

Country Link
CN (1) CN109376901B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110443430B (en) * 2019-08-13 2023-08-22 汕头大学 Block chain-based service quality prediction method
CN110766166B (en) * 2019-10-23 2021-03-23 支付宝(杭州)信息技术有限公司 Push model optimization method and device executed by user terminal
CN111541570B (en) * 2020-04-22 2021-05-07 北京交通大学 Cloud service QoS prediction method based on multi-source feature learning
CN112600697B (en) * 2020-12-07 2023-03-14 中山大学 QoS prediction method and system based on federal learning, client and server

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104778636A (en) * 2015-04-08 2015-07-15 中国人民解放军海军工程大学 Ship equipment maintenance support information service recommendation method
CN104915444A (en) * 2015-06-29 2015-09-16 北京邮电大学 Information recommending method and device
CN105024886A (en) * 2015-07-31 2015-11-04 浙江大学 Rapid Web server QoS (Quality of Service) prediction method based on user metadata
CN105791010A (en) * 2016-03-07 2016-07-20 天津职业技术师范大学 Quality-of-service (QoS) predicting method based on user implicit characteristics and collaborative filtering

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104778636A (en) * 2015-04-08 2015-07-15 中国人民解放军海军工程大学 Ship equipment maintenance support information service recommendation method
CN104915444A (en) * 2015-06-29 2015-09-16 北京邮电大学 Information recommending method and device
CN105024886A (en) * 2015-07-31 2015-11-04 浙江大学 Rapid Web server QoS (Quality of Service) prediction method based on user metadata
CN105791010A (en) * 2016-03-07 2016-07-20 天津职业技术师范大学 Quality-of-service (QoS) predicting method based on user implicit characteristics and collaborative filtering

Also Published As

Publication number Publication date
CN109376901A (en) 2019-02-22

Similar Documents

Publication Publication Date Title
CN109376901B (en) Service quality prediction method based on decentralized matrix decomposition
Wu Cloud-edge orchestration for the Internet of Things: Architecture and AI-powered data processing
Du et al. Federated learning for vehicular internet of things: Recent advances and open issues
Huang et al. Distributed reputation management for secure and efficient vehicular edge computing and networks
Jayasinghe et al. Rpr: A trust computation model for social internet of things
Caminha et al. A smart trust management method to detect on-off attacks in the internet of things
EP2498440B1 (en) Configuration method and system of complex network and configuration and management module of server resources
Malik et al. Vehicular networks with security and trust management solutions: proposed secured message exchange via blockchain technology
Saputra et al. Dynamic federated learning-based economic framework for internet-of-vehicles
Fu et al. GRAP: Grey risk assessment based on projection in ad hoc networks
Chang et al. Protecting mobile crowd sensing against sybil attacks using cloud based trust management system
CN112910710A (en) Network flow space-time prediction method and device, computer equipment and storage medium
CN114556381A (en) Developing machine learning models
Rathee et al. A trusted social network using hypothetical mathematical model and decision-based scheme
Ramadan et al. A memetic optimization algorithm for multi-constrained multicast routing in ad hoc networks
CN109195098B (en) Shared collaborative filtering method based on differential privacy
Esmaeilyfard et al. A privacy‐preserving mechanism for social mobile crowdsensing using game theory
Luo et al. On the universality of the Jordan center for estimating the rumor source in a social network
Sirisala et al. A novel trust recommendation model in online social networks using soft computing methods
CN113766506B (en) Hierarchical access control method for Internet of things
CN116488906A (en) Safe and efficient model co-building method
Wang et al. UbiPaPaGo: Context-aware path planning
El Husseini et al. Trust-based authentication scheme with user rating for low-resource devices in smart environments
CN114969465A (en) Federal learning data processing method, device and equipment
CN113157938A (en) Method and device for jointly processing multiple knowledge graphs for protecting privacy data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant