Web Service QoS Prediction Approach in Mobile Internet Environments

Web Service QoS Prediction Approach in Mobile Internet Environments
Lubao Wang, Qibo Sun, Shangguang Wang*, You Ma, Jinliang Xu, Jinglin Li
State Key Laboratory of Networking and Switching Technology
Beijing University of Posts and Telecommunications
Haidian, Beijing, 10086 China
imbarainbao@gmail.com; {qbsun; sgwang; youma; jlxu; jlin}@bupt.edu.cn
Abstract—Existing many Web service QoS prediction
approaches are very accurate in Internet environments,
however they cannot provide accurate prediction values in
Mobile Internet environments since QoS values of Web
services have great volatility. In this paper, we propose an
accurate Web service QoS prediction approach by weakening
the volatility of QoS data from Web services in Mobile Internet
environments. This approach contains three process, i.e., QoS
preprocessing, user similarity computing, and QoS predicting.
We have implemented our proposed approach with experiment
based on real world and synthetic datasets. The results show
that our approach outperforms other approaches in Mobile
Internet environments.
Keywords- Web services; QoS; collaborative filtering;
correlation coefficient
I. INTRODUCTION
With the rapid development of Mobile Internet, a large
number of Web services had emerged. Then it is very
difficult to select suitable Web services from these services
with the same function but different Quality of Services
(QoS). Users want to know which services is better,
especially QoS. Hence, how to accurately predict the QoS
values of each Web services before users use these services
is a very important issue.[1]
If you want to accurately choose the most appropriate
Web service to meet user demand. You need to make an
accurate prediction for QoS, and choose the best one. But,
the existing approaches will have large prediction error
when using these approaches to predict QoS values in
Mobile Internet environments. The main reason is that
compared to the traditional Internet, QoS values of Web
services from Mobile Internet have greater volatility.
II. OUR APPROACH
In order to avoid the volatility of QoS values, we
proposed a Web service QoS prediction approach (called
WSQP) by weakening the volatility of QoS data from Web
services in Mobile Internet environments. This approach
first uses a preprocessing strategy to reduce the volatility of
QoS values. Then we adopt Pearson Correlation Coefficient
(PCC) to find similar users. Finally, we predict the QoS
values by using normal QoS data that can be obtain from all
similar users history QoS data.
A QoS preprocessing strategy
When Web services runs in Mobile Internet, Packet loss,
delay, retransmission phenomenon is a common occurrence.
So QoS values of the Mobile Internet have greater volatility,
and there are many normal values and abnormal values in
history QoS data. Although the number of abnormal values
is very small, they often degrade the accuracy of QoS
prediction for Web services. Hence, it is very essential to
weaken the abnormal QoS values of Web services.
In this paper, we assume ,
t
a j
q represents the history
QoS value of user a repeatedly invokes Web service
j(j=1,2,3,...) for the t(t=1,2,3,...) time .Then the history QoS
data set of user a repeatedly t times invokes Web service j
is 1 2 3
, , , , ,
{ , , , , }
history t
a j a j a j a j a j
Q q q q q }
t
a j,
,,, . In order to weak the volatility
of QoS data from Web services in Mobile Internet
environments, a strategy is used as follows:
, ,ln( )t t
a j a jp q (1)
Where ,
t
a j
p represents the preprocessed result of the
history QoS data ,
t
a j
q ; t represents the number of repeated
times. The history QoS data set of user a repeatedly t times
invokes Web service j can be transformed a set weakens the
volatility, i.e., 1 2 3
, , , , ,
{ , , , , }
t
a j a j a j a j a j
Q p p p p, }
t
a j,
,,, .
B User similarity computation
Pearson Correlation Coefficient (PCC) has been
introduced in a number of recommender systems for
similarity computation, since it can be easily implemented
and can achieve high accuracy. PCC is employed to calculate
the similarity between two service users a and u using the
following equation:
, , , ,
1
,
1 , ,
1
( )( )
1
t
k k
a l a l u l u ln
k
a u
l a l u l
p E p E
t
sim
n D D
¦
¦ (2)
Where ,a usim represents the similarity between two
service users a and u ; , , ,
1
1
a l
t
k
a l a l
k
E E p
t
¦ represents the
average value of ,a l
Q ; , , ,
1
1 t
k
u l u l a l
k
E E p
t
¦ represents the
average value of ,u l
Q ; 2
, , , ,
1
1
( )
t
k
a l a l a l a l
k
D D p E
t
¦ represents
the variance of ,a l
Q ; 2
, , , ,
1
1
( )
t
k
u l u l u l u l
k
D D p E
t
¦ represents the
variance of ,u l
Q .
Although PCC can provide accurate similarity
computation, it will overestimate the similarities of users
who are actually not similar but happen to have similar QoS
experience on a few common invoked Web services [2]. To
2014 IEEE International Conference on Data Mining Workshop
978-1-4799-4274-9/14 $31.00 © 2014 IEEE
DOI 10.1109/ICDMW.2014.27
1239

address this problem, we employ a significance weight to
reduce the influence of a small number of similar common
invoked services. An enhanced PCC for the similarity
computation between different users is defined as
, ,
| |
| | | |
a u
a u a u
a u
US US
sim sim
US US
ˆ
c
u
(3)
Where ,a usimc represents the new similarity value;
represents the number of Web service items that are
employed by both the two users; | |a
US and are the number
of Web services invoked by user a and user u respectively.
C QoS prediction
Based on the similarity between every two users, if we
want to predict the QoS value of user a invoke Web service j,
we need to find the most similar users with user a and select
a normal values interval based on the characteristics of the
QoS data. Then using the following equations:
,
1
1
, c
v
e
c u j
e
F u j q
v
¦ (4)
,
1 1
,
1
1
( ( , ) - ( , ))c
c
K K
a u c c
c c
n
a u
c
wight
sim F u j F u j
K
sim
F
c
c
¦ ¦
¦
(5)
1
1
( , )
K
c
c
wightForecast F u j F
K
¦ (6)
Where Forecast represents the prediction QoS value of
user a access Web service j; ,c
e
u j
q represents a QoS value
from ,c
u j
Q which is in this normal values interval; v represents
the number of QoS values which are in the normal values
interval; K represents the number of most similar users with
user a; (1 )cu c Kd d represents a user who is similar with
user a.
The normal values interval which is selected by the
characteristics of QoS data only conclude the normal data.
For example, the data of response time tends to become large
when affected, and the data of throughput tends to become
smaller when affected. So for the response time data, the
normal values interval should be min ,
( )
c
u j
p E ( ,c
u j
E represents
the average value of ,c
u j
Q ,i.e., , ,
1
1
c c
t
k
u j u j
k
E p
t
¦ ), for the
throughput, the normal values interval should be... min
p is the
smallest value in ,c
u j
Q which is the result QoS set of
preprocessed the history QoS set of user cu repeatedly t
times access Web service j. max
p is the largest value in ,c
u j
Q .
III. EXPERIMENTS
We implement our approach 1
and conduct experiments
1
http://www.sguangwang.com/Source code/wang-demo.avi
using a datasets 2
named WSDream. It contains nearly
1million service response time records. We conduct
experiments to compare WSQP (our approach) against
WSRec[3] and AVG (It`s based on the average of history
QoS data ) in terms of response time.
A. Comparison results on relative error
In this section, we perform experiments to compare
WSQP with other approaches in terms of relative error (RE)
where RE can be calculated as follows:
realRE F F (7)
Where F is the predicted QoS value; real
F is the real
value which we had deleted before predicting QoS value;
the smaller the absolute value of RE is, the more accurate
the prediction is.
As shown in Fig.1 (a), Fig.1 (b) and Fig.1 (c), we find
that comparing with WSRec and AVG, our approach is the
best.
(a)WSQP
(b)WSRec
2
http://www.wsdream.net
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
-4
-3
-2
-1
0
1
2
3
4
5
Prediction Times
RE
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
-5
-4
-3
-2
-1
0
1
2
3
4
5
Prediction Times
RE
1240

(c)AVG
Figure 1. Comparison WSQP with WSRec and AVG
B. Comparision results on mean absolute error
We use Mean Absolute Error (MAE) to evaluate the
prediction quality .MAE can be calculated as follows:
, ,
,
a j a j
reala j
F F
MAE
N
¦ (8)
Where
,a j
F represents the predicted QoS value of user a
invoked Web service j;
,a j
realF is the real value of user a in
voke Web service j which we had deleted before predicting
QoS value; N is the number of predicted values. The smaller
the absolute value of MAE is, the more accurate the predictio
n is.
As shown in Fig.2 (N=100), we find that the MAE value
of WSQP is the smallest in all approaches. These means our
approach can significantly improve the accuracy of QoS pre
diction for Web services in Mobile Internet environments.
Figure 2. Comparision results on MAE
IV. CONCLUSIONS
In this paper, we presented an easy and accuracy QoS
prediction approach for Web services in Mobile Internet
environments. The experimental results show that our
approach obtains more accuracy QoS prediction than other
approaches.
ACKNOWLEDGEMENTS
The work presented in this study is supported by the
Natural Science Foundation of Beijing under Grant
No.4132048, NSFC(61472047), and NSFC(61202435).
REFERENCES
[1] S. G. Wang, Q. B. Sun, and F. C. Yang, "Towards Web Service
selection based on QoS estimation," International Journal of Web and
Grid Services, 4, vol. 6, pp. 424-443, 2010.
[2] M.R. McLaughlin and J.L. Herlocker, “A Collaborative Filtering
Algorithm and Evaluation Metric that Accurately Model the User
Experience,” Proc. 27th Int’l ACM SIGIR Conf. Research and
Development in Information Retrieval (SIGIR ’04), pp. 329-336,
2004.
[3] Z. Zheng, H. Ma, M.R. Lyu, and I. King, “QoS-Aware Web Service
Recommendation by Collaborative Filtering,” IEEE Transactions
Service Computing, vol. 4, no. 2, pp. 140-152, 2011.
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
-15
-10
-5
0
5
10
15
20
Prediction Times
RE
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
0
0.5
1
1.5
2
2.5
3
3.5
Prediction Times
MAE
WSQP
WSRec
AVG
1241

Web Service QoS Prediction Approach in Mobile Internet Environments

More Related Content

What's hot (19)

Viewers also liked (13)

Similar to Web Service QoS Prediction Approach in Mobile Internet Environments (20)

More from jins0618 (20)

Recently uploaded (20)

Web Service QoS Prediction Approach in Mobile Internet Environments