Testing and evaluation for Web usability based on ... - Springer Link

3 downloads 123 Views 638KB Size Report
Web applications, it is unlikely that the application developers can enumerably validate all possible navigation paths to ensure the consistent and correct ...
WUJH$

Vol. 9 No. 5 2004 687-693

Wuhan University Journal of Natural Sciences

Article ID:1007 1202(2004)05-0687 07

Testing and Evaluation for Web Usability Based on Extended Markov Chain Model 0

Introduction

[] MAO Cheng-ying, LU Yan-sheng t College of Computer Science and Technology, Huazhong University of Science and Technology, Wuhan ,130074, Hubei, China

Abstract. As the increasing popularity and complexity of W{,b applications and the emergence of their new characteristics, the testing and maintenance of large, complex Web applications are becoming more complex and difficult. Web ap plications generally contain lots of pages and are used by enor mous users. Statistical testing is an effective way of ensuring their quality. Web usage can be accurately described by Markov chain which has been proved to be an ideal model for software statistical testing. The rcsuhs of unit testing can bc utilized in lhe tatter stages, which is an important strategy for ])otton~to top integration testing, and the other improvement {}f extended Markov chain model (F.MM) is to present the err(}r type vector which is treated as a part of page node. This paper also proposes the algorithm for generating test cases of usage paths. Finally, optional usage reliability evaluation mcthods and an incremental usability regression testing model for testing and evaluation are presemed.

Key words: statistical testing; evaluati(m for Web usability; extended Markov chain model ( E M M ) : Web log mining: reliability evaluation ct.c number. "IP311.5

Received date= 2004-06-15 Foundation item: Supported by the National l)cfcnce Research Pro-

ject ( No. 11315. 9. 2) and Naional Science and "Ibchnology Plan ( 2001 BA1 o2A04 02-03) Biography= MA{} Chcng-ying ( 1978 ), male. Ph.D. candidate, research dircction: software testing/. Rcs{~ar('hdirecti{m: advanced database system, s{}ftwarc testmI4. COmlXmCnl teclmo[ogy and data mining. E m~tit: maocl~y(a:yeal~,net 1" "Ib whom correspon{k,nce sh{}uld be addressed

ecently, with the enormous evolution and prevalent utiIization of network technology, Web applications are rapidly diffused for their unique merits, such as numerous concurrent users, interaction, functionality distribution and facility. Meanwhile, more and more strict requirements of usability, reliability and security are put forward. Due to the characteristics of distribution, heterogeneity and document focus, it is hard to get the information about control flow and data flow, therefore methods and techniques for testing and ensuring the reliability of traditional software can't be used for Web applications. Presently, there are some explorative re search results on the Web applications testing, but it is still at the embryonic stageEI3. We can adopt the full grown methods of structural and functional testing to implement the page test (also called unit test). But in the phase of latter integration testing, due to nu merous pages and complex dependences between them in most Web applications, it is unlikely that the application developers can enumerably validate all possible navigation paths to ensure the consistent and correct information flow. But testing for the navigation behaviors is very necessary, because users' estimate of a Web application and judgement of its reliability are directly from the percentage of hitting the target pages in the access process, i.e. usability of Web application. Statistical testing~e'3l , also called random testing, executes the testing with a sample from the test case space firstly, then fixes the faults and evaluates the reliability of Web applications. In contrast to other methods of testing, this approach has the benefits including greatly reducing testing effort, providing an appropriate estimate for the software reliability, etc. ()bviously,

R

687

statistical testing is very suitable for the testing of Web usability. Markov chain is a model of ideal statistical testing of the softwareEel. For Web applications, we can gain better effect by adopting the Markov chain to check links and test their usageE3~2. Li et al affirmatively answered the question-" Can Web usage be accurately modeled by Markov chain.'?", based on the experiments on real Web sites E6~. Furthermore, the research results of Zhu et al also indicate that Markov chain is very suitable for modeling Web users' navigation on the WWW Er~. In this paper, we import the error type vector to ex tend Markov chain model based on their previous works, and ultimately propose a semi-automatic statistical testing and evaluation model for Web usage.

1 Preliminary Knowledge 1.1 Charactersof Web Applications Web application architecture commonly consists of three major tiers. The client browser and Web server harmoniously interact by means of the request/response HTTP protocol, especially in newly distributed applications, client objects connect to server objects by means of some underlying object framework like CORBA. Web application can be viewed as the aggregation of a series of pages and other components that are used to carry out some tasks, and the relations and navigation behaviors are produced by linksE&

cdgc coverage(E(') of the "I'C; &.8, N('~~,, 8 . c~.. E { ' : : ~ )

18

order lhe paths in T C by 1.he priority of usage probability;

19

rot urn T C ;

2O }

"Ihble 2 gives the coverage of partial test cases of the sample t)rogram. And the trend of coverage rate increas ing with the number of test cases is demonslrated in Fig. 2. After gaining test cases, the testing can be driven by a script executor automatically if the test cases are ex pressed in a type of script files II , also can be manually executed. The 10'h of the types of errors in "Ihble 1 can be filled with lester's evaluation of page's functionality. such as the validity of page's output content resulted from the input, the layout of components etc. During the regression testing, the exit probability of a page the transition probability from the page to virtual exit t)age- can be estimated more accurately by means of mining sequential patterns from the former logs and sub sequently counting the time number of the page being viewed as terminal page.

3.2 Reliability Evaluation of Usage The purpose of testing is to detect the latent faults "lhble 2

Coverage of the first ten test eases of the sample

No. "I\!st path

U{.7

%

Aceumulalivc Accumulative Accumulalive UC

NC

E{"

1

0 1e

15.00

13.00

18.75

1. 17

2

0 1e

10.00

25.00

2:5.00

8.33

3

0:5 e

8.25

33.25

31.25

12.50

1

0-I 1 e

6.00

39.25

31.25

11.58

5

0 13 e

5.00

lI. 25

s

18.75

6

0 3 e

5.00

19.25

37.50

20.83

7

0 15 e

l. 13

53.38

37.50

22. !)'2

8

0 1 8 e

3.75

57.13

I;L 75

27.08

9

0 .t 7 e

3. s

60.68

50.00

31.25

2.70

63.33

56.25

35. 12

10 0 5 l0 e

10 I

0

express the usage probabilily as lhe product of lransltion

13

3O 20

5

Fig. 2

VoL 9

No. 5

2004

20 25 30 35 Number of test cases

4O

45

50

Coverage versus number of test cases

in applicalions and provide reasonable reliability assessmenl anti prediction. After driving the statistical testing by the model of EMM, the access logs and error logs are important guidance to fix the faults and improve the qualhy of the applicalions. Firstly. we conduct data cleaning on the two logs in ways of removing the requests of services to the multimedia such as picture and video without using GET protocol, deleting the error records caused by ISP in the error logs. Secondly. count out the number of request for each l)age, i.e. U N i ( I ~ i ~ [ N I ) , and compute the sum of them by the formula: N

UN = ~

UNi

Meanwhile, the error type vector el/----~e~ ,"" , ~ ! ! ~setnng../ I .

I documcntsk]J

=t

c _ ~

II

' . 114. . . . .

~

~ ,,~

I Cr

' I

~

~

~

"\

I I type~ t o ~ l /

~

'~

\'/

iI.,.~ -- ~testin.K r -- - V, -incremental = - -- =- improve[,,. -_-=- -_-........ ~ , , the I~reliabilit'~l . . . . . . . ,..] evaluate

........... Fig. 3

/

reliability I - ~PL..m.~ Ja

Model for Web usage testing and evaluation based on E M M

As depicted in the above flow chart, our proposed model is a semi-automatic framework for Web usage tes ring and reliability evaluation. The manual interventions required to reverse engineer and test a Web application are indicated within diamonds. Other parts can be automatically executed by small programs which are developed by ourselves or improved based on existing tools. Moreover, there are many frequent changes of users' de mands and strictly fixed deadline in the most Web applications, so the incremental regression testing strategy of the above model is very necessary. Under the background of developing programs in B/S pattern of the digital university project of HUST, we are actively constructing a comprehensive incremental regression testing platform for Web usage testing and quality assurance. 692

L - - the E M~__ - - [ " -

4

Conclusion

Compared with the traditional software, Web applications have special properties, such as numerous users and Web pages, loose coupling of pages, etc. To validate and evaluate Web usability, a non-traditional approach (statistical usage testing) is employed. It has many benefits such as: high efficiency, low cost and quantitative analysis. Since the clients are final users of Web applications, their opinions are crucial factors for assessing qual ity of the applications Els3. So the statistical usage testing is a new applied strategy for quality assurance and improvement of Web applications. The proposed EMM smoothly absorbs the results of unit testing (or the relia hility of a page), and has been enriched with the error

type vector to simplify the work of fixing faults and facilitate the improvement of EMM during regression testing. Based on the EMM, we also provide an algorithm for generating test paths. Finally, according to the styles of testing execution (automatically or manually), the accu rate or approximate assessment scheme has been presen ted. And an incremental usage testing model integrated with testing and evaluation has come into practice. A critical review of our method highlights some di rections for future research. Except that the virtual exit page and exit probability are introduced to terminate test paths, we can also mine the distribution of users' browse steps from access logs, and then generate random number of that distribution as the length of test paths. It is necessary for us to further wtlidate the proposed evaluation model in more large-scale and complex Web applications and to deploy it associated with Web log mining techniques to reap the full benefit of ensuring and improving the quality of Web applications. Furthermore, applying the testing and evaluation model based on EMM to the development of Web application is also a direction of our next researches.

References ~1~

~2~

~3~

Miller E. The Weh Site Quality Challenge. Proceedi~g . f Q W ' 9 8 , l lth Annual Imernatim~a! Swen, Chen Zhe~>@lng. A Survey of Web Test ing. ( "oreputer Sciem'e, 2003,30( 3 ) : 100 104 (('h). ~4~

[]

693