• The Journal of the Korea Contents Association
• /
• v.9 no.9
• /
• pp.80-89
• /
• 2009

A fast and efficient search algorithm to determine an optimal number of clusters in clustering algorithms is presented. The method is based on cluster validity index which is a measure for clustering optimality. As the clustering procedure progresses and reaches an optimal cluster configuration, the cluster validity index is expected to be minimized or maximized. In this Paper, a fast non-exhaustive search method for finding the optimal number of clusters is designed and shown to work well in clustering. The proposed algorithm is implemented with the k-mean++ algorithm as underlying clustering techniques using CB and PBM as a cluster validity index. Experimental results show that the proposed method provides the computation time efficiency without loss of accuracy on several artificial and real-life data sets.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2004.05a
• /
• pp.673-676
• /
• 2004

Fast and high-quality document clustering algorithms play an important role in providing data exploration by organizing large amounts of information into a small number of meaningful clusters. In particular, hierarchical clustering provide a view of the data at different levels, making the large document collections are adapted to people's instinctive and interested requires. Many papers have shown that the hierarchical clustering method takes good-performance, but is limited because of its quadratic time complexity. In contrast, K-means has a time complexity that is linear in the number of documents, but is thought to produce inferior clusters. Think of the factor of simpleness, high-quality and high-efficiency, we combine the two approaches providing a new system named CONDOR system [10] with hierarchical structure based on document clustering using K-means algorithm to "get the best of both worlds". The performance of CONDOR system is compared with the VIVISIMO hierarchical clustering system [9], and performance is analyzed on feature words selection of specific topics and the optimum hierarchy depth.

• Journal of Haehwa Medicine
• /
• v.23 no.1
• /
• pp.115-124
• /
• 2014

Objectives : There are plenty of medical record of acupuncture & moxibustion in Traditional East Asian medicine(TEAM). We performed this study to find out the hidden criteria lies on this record to choose proper acupoints. Methods : "Zhenjiuzishengjing", ancient TEAM book was analysed using document clustering techniques. Corpus was made from this book. It contained 196 texts driven from each symptoms. Each texts converted to vector representing frequency of 349 acupoints. Distance of vectors calculated by weighted Euclidean distance method. According to this distances, hierarchical clustering of symptoms was builded. Results : The cluster consisted of five large groups. they had high corelation with body part; head and face, chest, abdomen, upper extremity, lower extremity, back. Conclusions : It assumes that body part of symptom is the most importance criteria of acupoints selecting. some high similar symptom vectors consolidated this result. the other criteria is cause and pathway of illness. some symptoms bound together which had common cause and pathway.

• Phonetics and Speech Sciences
• /
• v.6 no.1
• /
• pp.77-83
• /
• 2014

This paper proposes a speaker and environment clustering method in order to overcome the degradation of the speech recognition performance caused by various noise and speaker characteristics. In this paper, instead of using the distance between Gaussian mixture model (GMM) weight vectors as in the Google's approach, the distance between the adapted mean vectors based on the modified maximum a posteriori (MAP) adaptation is used as a distance measure for vector quantization (VQ) clustering. According to our experiments on the simulation data generated by adding noise to clean speech, the proposed clustering method yields error rate reduction of 10.6% compared with baseline speaker-independent (SI) model, which is slightly better performance than the Google's approach.

• The KIPS Transactions:PartB
• /
• v.17B no.2
• /
• pp.177-182
• /
• 2010

This paper proposes a new document clustering method using PCA and fuzzy association. The proposed method can represent an inherent structure of document clusters better since it select the cluster label and terms of representing cluster by semantic features based on PCA. Also it can improve the quality of document clustering because the clustered documents by using fuzzy association values distinguish well dissimilar documents in clusters. The experimental results demonstrate that the proposed method achieves better performance than other document clustering methods.

• Proceedings of the KIEE Conference
• /
• 2006.11a
• /
• pp.225-227
• /
• 2006

The transmission network clustering using a fuzzy entropy function are proposed in this paper. We can define a similarity measure through a fuzzy entropy. All node in the transmission network system has its own values indicating the physical characteristics of that system and the similarity measure in this paper is defined through the system-wide characteristic values at each node. However, to tackle the geometric mis-clustering problem, that is, to avoid the clustering of geometrically distant locations with similar measures, the locational informations are properly considered and incorporated in the proposed similarity measure. In this paper, a new regional clustering measure for the transmission network system is proposed and proved. The proposed measure is verified through IEEE 39 bus system.

• Genomics & Informatics
• /
• v.1 no.1
• /
• pp.32-39
• /
• 2003

We present a latent variable model-based approach to the analysis of gene expression patterns, coupled with topographic clustering. Aspect model, a latent variable model for dyadic data, is applied to extract latent patterns underlying complex variations of gene expression levels. Then a topographic clustering is performed to find coherent groups of genes, based on the extracted latent patterns as well as individual gene expression behaviors. Applied to cell cycle­regulated genes of the yeast Saccharomyces cerevisiae, the proposed method could discover biologically meaningful patterns related with characteristic expression behavior in particular cell cycle phases. In addition, the display of the variation in the composition of these latent patterns on the cluster map provided more facilitated interpretation of the resulting cluster structure. From this, we argue that latent variable models, coupled with topographic clustering, are a promising tool for explorative analysis of gene expression data.

• Proceedings of the IEEK Conference
• /
• 2008.06a
• /
• pp.939-940
• /
• 2008

In this paper, we propose a novel image clustering method based on weighted sub-sampling to reduce clustering time and the number of clusters for target detection and tracking. Our proposed method first obtain sub-sampling image with specific weights which is the number of target pixels in sampling region. After performing clustering procedure, the cluster center position is properly obtained using weights of target pixels in the cluster. Therefore, our proposed method can not only reduce clustering time, but also obtain proper cluster center.

• Journal of the Korean Institute of Gas
• /
• v.4 no.4 s.12
• /
• pp.58-64
• /
• 2000

Data compression and retrieval method are investigated for the effective utilization of measured process data. In this paper, a new data compression method, Clustering Compression(CC), which is based on the k-means clustering algorithm and piecewise linear approximation method is suggested. Case studies on industrial data set showed the superior performance of clustering based techniques compared to other conventional methods and showed that CC could handle the compression of multi-dimensional data.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.5 no.2
• /
• pp.95-101
• /
• 2005

In this paper, we proposed a learning algorithm for the neuro-fuzzy modeling using a learning rule to adapt clustering. The proposed algorithm includes the data partition, assigning the rule into the process of partition, and optimizing the parameters using predetermined threshold value in self-constructing algorithm. In order to improve the clustering, the learning method of neuro-fuzzy model is extended and the learning scheme has been modified such that the learning of overall model is extended based on the error-derivative learning. The effect of the proposed method is presented using simulation compare with previous ones.