Effectiveness Of Clustering Based Interactive System

Published Date: 02 Nov 2017

Abstract- Interaction is very important for any information system. The quality of the product can be predicted by the user feedback. To improve the quality of service on the developer side, the developer needs to understand the user goals by analyzing the survey. The user goals, needs are to be captured with design tools. Qualitative and Quantitative clustering technique plays a major role in grouping the users based on their needs. The existing qualitative method becomes difficult as the number of users gets increased. Therefore the quantitative clustering techniques are also used to evaluate the goals. A method based on quantitative technique called Principal Component Analysis(PCA) is a data reduction technique, which is based on correlation. An automatic non-parameter Uncorrelated Discriminant Analysis(UDA) algorithm, satisfies the statistical uncorrelation property. The outcome of the framework yields users goals to the developer community. Based on that, the developer would develop a product which is consensus with the user needs.

Index Terms-Clustering, Interactive system, user centered design

I. INTRODUCTION

The failure of many information system is due to the lack of interaction between developer and the user. For IS development and design approach, there are various problem that may occur[1]. The beginning of the product or application has struggled with the challenges of communication between the users and the developer. As organization rely more on digital capital to drive innovation and achieve competitive advantages in worldwide market[2], it is more important than ever to understand the antecedents of system success and avoid the cause of failure. System use and success mainly focus on Information System design.

Interaction is very important for the Information System. For the improvement of product quality, the major criteria is user satisfaction and system success. The user satisfaction can be known from the user feedback. User-Centered Process Framework Focus on the user needs and goals of the product. This framework contains the user feedback in the form of Qualitative and Quantitative data. To improve the quality of service on the developer side, the developed needs to understand the user goals by analyzing the survey. Qualitative and Quantitative clustering technique plays a major role in grouping the user based on their needs[3]. These technique include manual and semi-automated methods Manual method require human judgement to identify users with similar characteristics. Semi-automated method based on the transaction log or statistical software. User centered process framework group the users by using these clustering technique[4]. The existing qualitative method becomes difficult as the number of users gets increased. To evaluate the user goals efficiently for large number of users quantitative method is used. The data that can be collected by qualitative and quantitative is as follows

Qualitative data can be collected through the interviews and online activity

Quantitative data can be collected through the numeric surveys and usage statistics

Based on these data, clustering technique is performed. The outcome of the framework yields the users goals to the developer community. Based on the outcome, the developer would develop a product which is consenus with the user needs.

II. RELATED WORK

A. User-Centered Design

User-centered design(UCD) is a type of interface design and a process in which the needs, wants, and limitation of the end users of a product are given extensive attention at each stage of the design process[4],[17].

User-centered design can be characterized as a multi-stage problem solving process that not only requires designers to analyze and foresee how users are likely to use a product, but also to test the validity of their assumptions with regards to user behavior in the real world tests with actual user. Such testing is necessary as it is often very difficult for the designers of a product to understand intuitively what a first-time user of their design experiences, and what each user’s learning curve may look like

The difference from user-centered design from product design philosophies that tries to optimize the product around how users can, want, or need to use the product, rather than forcing the user to change their behavior to accommodate the product. User-centered design process can help the software designers to fulfill the goal of a product engineered for their users. User feedback are considered right from the beginning and included into the whole product cycle

B. Quantitative Clustering Method

Principal component Analysis (PCA)[7], [8] , [11] is a quantitative clustering method. Understanding user information needs and mental models is important for design in information-rich domains. Information architects use card-sorting and other methods to understand user mental models for better design. In existing system persona development processes emphasize precision, but not accuracy. [17]The designer makes a subjective judgment regarding what user archetypes to focus on, a judgment that might be difficult for inexperienced designers. Even for experienced designers, personas based on the same user research might vary widely, because there is no tight coupling between user [7]. Finally, persona development relies mostly on interviews and observation, techniques that are useful for gaining deeper insight into a few users, but are not economical for gaining a broader understanding of target user groups[13]. The goal is to create a tighter coupling between user research and persona development by using quantitative methods to identify types of information needs.

It uses Principal Components Analysis (PCA)[14], an exploratory data analysis technique that can reduce the dimensionality of large datasets, by identifying important underlying factors.

Method

Step 1: Get some data

Step 2: Subtract the mean

For calculating mean

X/n

Step 3: Calculate the covariance matrix

Step 4: Calculate the eigenvectors and eigenvalues of the covariance matrix

Step 5: Choosing the components and forming the feature vector

It reduce the original variables into new components that convey as much of the original data as possible.It is based on correlation.

C. Singular Value Decomposition

SVD is a method for identifying and ordering the dimensions along which data points exhibit the most variation.[18] These are the basic ideas behind SVD: taking a high dimensional, highly variable set of data points and reducing it to a lower dimensional space that exposes the substructure of the original data more clearly and orders it from most variation to the least[18]. SVD can simply ignore variation below a particular threshhold to massively reduce the data but be assured that the main relationships of interest have been preserved.

SVD is based on a theorem from linear algebra which says that a rectangular matrix A can be broken down into the product of three matrices - an orthogonal matrix U, a diagonal matrix S, and the transpose of an orthogonal matrix V . The theorem is usually presented like this:

Where ; ; the columns of U are orthonormal eigenvectors of , the columns of V are orthonormal eigenvectors of , and S is a diagonal matrix containing the square roots of eigenvalues from U or V in descending order.

III. PROPOSED WORK

Clustering Schemes

Uncorrelated Discriminant Analysis

Ucp Framework

Evaluation Engine

Principal Component Analysis

Fig.1 System Architecture

To improve the quality of service on the developer side, the developer needs to understand the user goals by analyzing the survey of User-Centered Process(UCP) Framework. The existing qualitative method becomes difficult as the number of users gets increased. Therefore the quantitative method of Principal Component Analysis(PCA) is a data reduction technique, which is based on correlation

That it only measure linear relationship between X & Y for any relationship to exist, any change in X has to have a constant proportional change in Y. If the relationship is not linear than the result is inaccurate.

To overcome this limitation Uncorrelated Discriminant Analysis (UDA) is used[9]. In this method, there is no parameter in the whole process and an entire automatic strategy is established. New test sample is also produced through this method. It gives the accurate result than PCA and it can be evaluated.

UDA algorithm based on maximum margin criterion (MMC). MMC is connected with a symmetric matrix, its eigen-vectors can be chosen mutually orthogonal. Feature extraction method based on MMC criterion is robust, stable, and efficient. The extracted features via UDA are statistically uncorrelated. UDA combines rank preserving dimensionality reduction and constraint discriminant analysis, and also serves as an effective solution for small sample size problem[9].

Assume that are c known pattern classes. Given a data matrix X = []Є , where each column of X denotes a training data point in the d-dimentional space. Suppose that and (i= 1,…,c) are the mean vectr and sample number of class , respectively, and that m is the total mean vector[9]. The between-class scatter matrix , the within-class scatter matrix , and the total scatter matrix are determined by the following formulas:

It is easy to verify that St=Sb+Sw Define matrics

Then the scatter matrices and can be expressed as and . Then is the centered matrix of X.

For the high dimensional data, d is very large. Hence, it is difficult to deal with the sample matrix directly. Therefore, it is necessary to find a linear projection from the high-dimensional space to a significantly low-dimensional feature space[9]. The singular value decomposition (SVD) is the well-known approach that can be used for dimensionality reduction.

Singular value decomposition (SVD) is a method for transforming correlated variables into a set of uncorrelated ones that better expose the various relationships among the original data items. At the same time, SVD is a method for identifying and ordering the dimensions along which data points exhibit the most variation. This ties in to the third way of viewing SVD, which is that once we have identied where the most variation is, it's possible to find the best approximation of the original data points using fewer dimensions[18]. Hence, SVD can be seen as a method for data reduction.

The UDA algorithm searches the statistical uncorrelated discriminant vectors based on MMC criterion in the subspace Φ. The first discriminant vector which is the eigenvector corresponding to the maximum eigenvalue of is obtained by:

UDA algorithm consist of the following four steps

Step 1: Calculate Hb or Ht from the training sample set X, and then compute the left singular matrix U of Ht or Hb by SVD. By the transformation U, the total, within-class and between-class scatter matrices in the lower-dimensional space become

Step 2: Compute the first discriminant vector which is the eigenvector corresponding to the maximum eigenvalue of .

Step 3: Compute the next discriminant vector ϕi according to Theorem 5, until the discriminant vector does not satisfy ϕi ∈ Φ. Finally, generate the linear transformation matrix Ψ and let T = UΨ.

Step 4: For a new test sample , the decision result is given by

UDA combines rank preserving dimensionality reduction and constrain discriminant analysis, it serve as effective solution for small sample size problem. It maximize the margin between the class. Therefore it produce the accurate result.

IV. EXPERIMENT RESULT

Data gathered through an online survey, containing both qualitative semistructure interview questions, standard quantitative numeric survey questions and server queries of system usage for an online survey. The ratings for the survey questions are:

User id

0

1

2

3

4

5

6

7

8

Ques 1

8

0

2

1

0

6

7

5

6

Ques 2

2

8

5

3

0

2

2

8

0

Trans 1

5

5

8

8

7

1

3

5

0

Trans 2

8

3

4

0

7

5

8

9

8

Goal 1

6

0

8

6

0

6

5

6

4

Goal 2

0

5

7

3

4

4

2

4

0

Behav1

4

2

2

5

7

7

0

9

8

Fea 1

5

5

1

9

0

2

7

8

1

Fea 2

7

8

7

8

2

0

5

6

1

Table.1 Survey Ratings

In PCA technique it cluster the users of same category[19]. In fig 2. PCA group the users into four categories. Based on correlation, it perform the clustering.It contains parameter.

PCA

CLUSTER 1

0,4,8,12,1,9,2,10,3,11,5,6,7,14

CLUSTER 2

13,21,25,29,20,27

CLUSTER 3

30,34,38,28,36,31,39,32,33,24,37,35

CLUSTER 4

15,19,23,18,16,17,26,22

Table.2 PCA Clustering

In UDA technique, it cluster the users of same category and doesnot contain any correlation between attributes[19]. Therefore the clustering is very accurate.

UDA

CLUSTER 1

1,5,9,13,2,10,3,11,4,12,6,7,8

CLUSTER 2

22,21,18,19,20,23,17,16,15,14

CLUSTER 3

27,31,26,28,25,29,24,30

CLUSTER 4

32,36,40,33,37,34,35,38,39

Table.3 UDA Clustering

The data collected by the above mentioned survey are clustered using both PCA and UDA. The performance evaluation shows that the UDA clustering gives accurate result than PCA. Due to uncorrelation and non-parameter strategy the result is more accurate

V. CONCLUSION

The user goals can be evaluated effectively through these techniques. The outcome of the framework yields users goals to the developer community. Based on that, the developer would develop a product which is in consensus with the user needs.

VI. REFERENCE

[1] M.A. Cook, Building Enterprise Information Architectures: Reengineering Information System. Prentice Hall, 1996

[2] V.Sambamurthy A.Bharadwaj, and V.Grover, "Shaping Agility through Digital Options: Reconcptualizing the Role of Information Technology in Contemporary Firms," MIS Quarterly, vol.27,no.2,pp.237-263,2003

[3] Brintrup, A.M., Ramsden.J, Takagi.H, Tiwari.A, "Ergonomic Chair Design by Fusing Qualitative and Quantitative Criteria Using Interactive Genetic Algorithms", 2008

[4] Larry L. Constantine, "Beyond User-Centered Design and Experience: Dsigning for User Performance", 2004

[5] T.Miaskiewicz, T.Sumner, and K.A.Kozar, "A Latent Semantic Analysis Methodology for the Identification and Creation of Personas," Proc. 26th Ann.SIGCHI conf.Human Factors in Computing Systems, pp.1501-1510,2008.

[6] T.K.Landauer, P.W.Foltz, and D.Laham, "An Introduction to Latent Semantic Analysis,"Discourse Processes,vol.25,pp.259-284,1998

[7] R.Sinha, "Persona Development for Information-Rich Domains,"Proc.Computer Human Interaction Extended Abstracts on Human Factors in Computing system(CHI EA’03), pp.830-831,2003.

[8] J.Pruitt and J.Grudin, "Personas: Practice and Theory,"Proc.Conf.Designing for User Experiences,pp.1-15,2003

[9] Wen-Hui Yang, Dao-Qing Dai, and Hong Yan, "Feature extraction and Uncorrelated discriminant analysis for high-dimensional data,"2007

[10] Y.Liang, C.Li,W.Gong and Y.Pan, "Uncorrelated Linear Discriminant Analysis Based on Weighted Pairwis Fisher Criterion,"Pattern Recognition, Vol.40,no.12, pp. 3606-3615,2007

[11] Jonalan Brickey, Steven Walczak, Tony Burgess, "Comparing Semi-Automated Clustering Methods for Persona Development",2012

[12] A. Cooper and R. Reimann, About Face 2.0: The Essentials of Interaction Design. Wiley Publishing, 2003.

[13] Blomquist, A & Arvola,"Personas in Action: Ethnography in an Interaction Design Team.NordiCHI ",2002

[14] J.F. Hair, W.C. Black, B.J. Babin,R.E.Anderson, and R.L. Tatham, Multivariate Data Analysis, sixth ed. Pearson Education, 2006

[15] A. K. Qin, P. N. Suganthan and M. Loog, "Uncorrelated Heteroscedastic Lda Based on the Weighted Pairwise Chernoff Criterion," Pattern Recognition, vol. 38, no. 4 pp. 613–616, 2005.

[16] F. Long, "Real or Imaginary: The Effectiveness of Using Personas in Product Design," Proc. Irish Ergonomics Soc. Ann. Conf., Irish Ergonomics Rev., L.W. O’Sullivan, ed., Irish Ergonomics Soc., pp. 1- 10, 2009.

[17] J.H. Gerlach and F.Y. Kuo, "Understanding Human-Computer Interaction for Information Systems Design," MIS Quarterly, vol. 15, no. 4, pp. 527-549, 1991

[18] Kirk Baker, "Singular Value Decomposition Tutorial", 2005

[19] A.Vidhyavani, A.k.Ilavarasi,"User-Centered process Framework For the Realization of Clustering Based Interactive System",2012