Keywords

Multi viewpoint based similarity measure, Data mining, text mining, single viewpoint 
INTRODUCTION

Data mining is the domain which deals with discovering knowledge from databases. One of the data mining techniques available is clustering. The clustering is used to group objects similar. Clustering is very popular algorithm that has utility in real time application development. Intelligent decision systems can be used to cluster the available data objects in order to make well informed decisions. Search engines like Google in the real world also use clustering mechanism to group documents so as to speed up query processing. Basically clustering is of two types namely partitional clustering and hierarchical clustering. In this paper the focus on the partitional clustering algorithm. With respect to clustering algorithms, the Kmeans is one of the algorithms that produce clusters [2]. Kmeans algorithm takes given dataset and number of clusters as input and generates clusters. It is used in many real world applications. For instance it is used in banking domain for credit card fraud detection. It takes all credit card transactions as input and generates three clusters such as low, medium and high. It does mean that it groups all transactions into the three types. It is known as making a model. Provided this mode, the new transactions are compared with low, medium and high. If they are nearer to one of the groups, the transactions is likely genuine. If the new transaction amount is beyond high with huge difference, then the algorithm suspects the transaction as fraud transaction an alerts banker and customer. Thus Kmeans is used in real applications. The quality of clusters made by a clustering algorithm depends on the quality of similarity measure. For instance Kmeans uses Euclidian distance [3], [4]. 
Clustering methods are of three types. They are hierarchical clustering, data partitioning and data grouping [5]. Hierarchical clustering focuses on making hierarchy of clusters. Partitioonal clustering focuses on making partitions while data grouping is used to make overlapping clusters. In all the above mentioned clustering approaches, the similarity measure is important. They use singleviewpoint for measuring similarity. However, multiple viewpoints can provide more qualitative assessment of similarity. This is the motivation in this project. We use multiviewpoint measure to achieve highly accurate clusters. We aim at achieving lowest intercluster similarity and highest intracluster similarity. The remainder of the paper is structured as follows. Section II provides literary survey. Section III focuses on multi viewpoint similarity based measure. Section IV presents algorithms. Section V provides experimental evaluation. Section VI provides experimental results while section VII concludes the paper. 
RELATED WORK

In many real world applications clustering documents is very essential. For instance web search engines must use the technique in order to make clusters and improve query processing. For many years the technique is being used. In case of document clustering, a document is treated as an object and objects which are similar are grouped together. The words in document are treated as terms. Each term can have frequency in the document. The term frequency is used in measuring similarity. A vector is built using term frequency and documents. Weighting schemes such as TFIDF (Term Frequency – Inverse Document Frequency) is used to denote document vectors. The document clustering is of many types as explored in literature. They include nonnegative matrix factorization [7], probabilistic based methods [8], and information theoretic coclustering. These approaches use different approaches to find similarity among documents. Euclidian distance is used by many techniques to find similaritymeasure. 


Jacquard coefficients make use of direction and magnitude that is not used by cosine similarity and ED measures. When the documents are presented in unit vectors, it is same as cosine similarity. There is comparison explored between Pearson correlation and Jacquard techniques in [14]. In [17] tree similarity is sued while in [18] phrase based similarity is used. However, they both depend on a common method known as Hierarchical Agglomerative Clustering. The lower side of these techniques is that they incur heavy computational cost. There are measures found for clustering XML documents as well. One such measure is explored in [19] named “Structural Similarity” which make difference between document and text clustering. This paper focuses on a new multiviewpoint based similarity measure. 
MULTIVIEW POINT BASED SIMILARITY MEASURE

In this paper multiviewpoint based similarity measure is explored. It makes use of multiple view points to determine similarity between documents. This will make more informative assessment of similarity while making clusters. It computes similarity between documents as follows. 

ALGORITHMS PROPOSED

Many algorithms have been proposed to work on multiviewpoint similarity measure. The procedure for similarity matrix is as shown in Listing 1. 

Algorithm 1 – Functionality for making similarity matrix 
As per the procedure in Algorithm 1, it specifies that dl and di are closer and the dl is also closer to di as per the multiviewpoint similarity measure. The Algorithm 2 shows the validation procedure. 

Algorithm 2 –Validation Procedure 
The final validity is calculated by averaging overall rows. It is as given in line 14, that when validation score is higher, then the similarity is higher and thus eligible for clustering. Fig. 1 shows the validity scores of multiviewpoint similarity and cosine similarity. 
As can be seen in fig. 1, Series 4 is related to klbMVS, series 3 corresponds to klbCS, and series 2 corresponds to reutors7 while series 1 corresponds to reutors 7 CS. As shown in fig. 1 performance of MVS is higher when compared to that of CS. 

Algorithm 3–Alg for incremental cluster 
Algorithm 3 is shows algorithm with two phases. They are initialization and refinement. Selecting k documents as seeds are known as initialization which making initial positions while the refinement makes much iteration to form best clusters. Each iteration in refinement phase visits n number of documents in randomized fashion. After that verification is done for each document. If the document contains similarity, it is moved to the cluster. When no documents are there the iterations come to an end. 
PERFORMANCE EVALUATION

For performance evaluation for two criterion functions such as Ir, and Iv associated to multiviewpoint similarity measure. Bench mark datasets have been used to test the efficiency of our approach. The results are shown in table 1. 
EXPERIMENTAL SETUP AND EVALUATION

Our algorithm is compared with other algorithms for performing evaluation. They include Mmeans, Cut Algorithm, Min Max graph EJ which is nothing but CLUTO’s graph with graphCS , extended Jacquard which is nothing but CLUTOs graph with SpkMeans, Cosine Similarity which is nothing but Spherical KMeans with Cosine Similarity, MVSC Iv which is nothing but the proposed approach with Iv criterion function and MVSC Ir which is nothing but the proposed approach with Ir criterion. The results are presented in the next section. 
RESULTS

The result of experiments are presented in fig. 2 and 3. It shows graphs with the results of all clustering algorithms for 20 benchmark datasets. The results are given in two different graphs. Each graph shows the experimental results of 10 datasets. 
As shown in fig. 2(a) and (b), the proposed approach performs better when compared with other algorithms. GraphEJ performed better only in some cases. Overall performance of MVSC Iv and MVSC Kr is far better than others. The effect of is also presented on the performance of MVSC Ir. 
THE EFFECT OF α ON THE PERFORMANCE OF MVSC IR

It is understood that the cluster size and balance have their impact on the methods of partitional clustering based on criterion functions. In terms of accuracy and NMI, FScore the assessment is done and the results are presented in fig. 3. 
As shown in fig. 3, MVSR Ir’s performance is within 5% with regard to any evaluation metric for best case. 
CONCLUSION

In this paper we consider measuring similarity between two documents from multipleview points. Similarity measures are used in document clustering. Clustering techniques depend on the similarity measure. All existing similarity measures are single viewpoint based. They cannot provide more informative assessment of similarity between objects. However, in this paper we explored the multi viewpoint based similarity measure which makes more informative assessment to group objects into different clusters. This new measure achieves highest level of similarity. The aim of the new measure is to achieve highest intracluster similarity and lowest intercluster similarity. We built a prototype application that shows the proof of concept. The results revealed that the proposed approach is capable of measuring similarity more accurately. 
Tables at a glance


Table 1 


Figures at a glance



Figure 1 
Figure 2a 




Figure 2b 
Figure 3 


References

 A. Ahmad and L. Dey, “A method to compute distance between two categorical values of same attribute in unsupervised learning for categorical data set,” Pattern Recognit. Lett., vol. 28, no. 1, pp. 110 – 118, 2007.
 A. Banerjee, I. Dhillon, J. Ghosh, and S. Sra, “Clustering on the unit hypersphere using von MisesFisher distributions,” J. Mach. Learn. Res., vol. 6, pp. 1345–1382, Sep 2005.
 I. Dhillon and D. Modha, “Concept decompositions for large sparse text data using clustering,” Mach. Learn., vol. 42, no. 12, pp. 143–175, Jan 2001.
 I. S. Dhillon, S. Mallela, and D. S. Modha, “Informationtheoretic coclustering,” in KDD, 2003, pp. 89–98.
 S. Flesca, G. Manco, E. Masciari, L. Pontieri, and A. Pugliese, “Fast detection of xml structural similarity,” IEEE Trans. on Knowl. And Data Eng., vol. 17, no. 2, pp. 160–175, 2005.
 I. Guyon, U. von Luxburg, and R. C. Williamson, “Clustering: Science or Art?” NIPS’09 Workshop on Clustering Theory, 2009.
 D. Ienco, R. G. Pensa, and sR.Meo, “Contextbased distance learning for categorical data clustering,” in Proc. of the 8th Int. Symp. IDA, 2009, pp. 83–94.
 Leo Wanner (2004). “Introduction to Clustering Techniques”. Available online at: http://www.iula.upf.edu/materials/040701wanner.pdf [viewed: 16 August 2012]
 C. D. Manning, P. Raghavan, and H. Sch ¨ utze, An Introduction to Information Retrieval. Press, Cambridge U., 2009.
 on webpage clustering,” in Proc. of the 17th National Conf. on Artif. Intell.: Workshop of Artif. Intell.for Web Search.AAAI, Jul. 2000, pp. 58–64.
 J. Shi and J. Malik, “Normalized cuts and image segmentation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 22, pp. 888–905, 2000.
 A. Strehl, J. Ghosh, and R. Mooney, “Impact of similarity measures.
 X. Wu, V. Kumar, J. Ross Quinlan, J. Ghosh, Q. Yang, H. Motoda, G. J. McLachlan, A. Ng, B. Liu, P. S. Yu, Z.H. Zhou, M. Steinbach, D. J. Hand, and D. Steinberg, “Top 10 algorithms in data mining,” Knowl.Inf.Syst., vol. 14, no. 1, pp. 1–37, 2007.
 W. Xu, X. Liu, and Y. Gong, “Document clustering based on nonnegative matrix factorization,” in SIGIR, 2003, pp. 267–273.
 H. Zha, X. He, C. H. Q. Ding, M. Gu, and H. D. Simon, “Spectral relaxation for kmeans clustering,” in NIPS, 2001, pp. 1057–1064.
 Y. Zhao and G. Karypis, “Empirical and theoretical comparisons of selected criterion functions for document clustering,” Mach. Learn., vol. 55, no. 3, pp. 311–331, Jun 2004.
 S. Zhong, “Efficient online spherical Kmeans clustering,” in IEEE IJCNN, 2005, pp. 3180–3185.
