Kwangcheol Shin

Self organizing sensor networks using intelligent clustering

Minimization of the number of cluster heads in a wireless sensor network is a very important problem to reduce channel contention and to improve the efficiency of the algorithm when executed at the level of cluster-heads. This paper proposes a Self Organizing Sensor (SOS) network based on an intelligent clustering algorithm which does not require many user defined parameters and random selection to form clusters like in Algorithm for Cluster Establishment (ACE) [2]. The proposed SOS algorithm is compared with ACE and the empirical results clearly illustrate that the SOS algorithm can reduce the number of cluster heads.

Fast clustering algorithm for information organization

This study deals with information organization for more efficient Internet document search and browsing results. As the appropriate algorithm for this purpose, this study proposes the heuristic algorithm, which functions similarly with the star clustering algorithm but performs a more efficient time complexity of O(kn), (k≪n) instead of O(n2) found in the star clustering algorithm. The proposed heuristic algorithm applies the cosine similarity and sets vectors composed of the most non-zero elements as the initial standard value. The algorithm is purported to execute the clustering procedure based on the concept vector and produce clusters for information organization in O(kn) period of time. In order to see how fast the proposed algorithm is in producing clusters for organizing information, the algorithm is tested on TIME and CLASSIC3 in comparison with the star clustering algorithm.

Improving kNN text categorization by removing outliers from training set

We show that excluding outliers from the training data significantly improves kNN classifier, which in this case performs about 10% better than the best know method—Centroid-based classifier. Outliers are the elements whose similarity to the centroid of the corresponding category is below a threshold.

Improving Information Retrieval in MEDLINE by Modulating MeSH Term Weights

MEDLINE is a widely used very large database of natural language medical data, mainly abstracts of research papers in medical domain. The documents in it are manually supplied with keywords from a controlled vocabulary, called MeSH terms. We show that (1) a vector space model-based retrieval system applied to the full text of the documents gives much better results than the Boolean model-based system supplied with MEDLINE, and (2) assigning greater weights to the MeSH terms than to the terms in the text of the documents provides even better results than the standard vector space model. The resulting system outperforms the retrieval system supplied with MEDLINE as much as 2.4 times.

Enhanced centroid-based classification technique by filtering outliers

Document clustering or unsupervised document classification has been used to enhance information retrieval Recently this has become an intense area of research due to its practical importance Outliers are the elements whose similarity to the centroid of the corresponding category is below some threshold value In this paper, we show that excluding outliers from the noisy training data significantly improves the performance of the centroid-based classifier which is the best known method The proposed method performs about 10% better than the centroid-based classifier.

Efficient Web Services Composition and Optimization Techniques

In this paper, we suggest a simple but efficient solution for the Web service composition. When we search Web services for composition, we visit the service which gives the biggest number of new responses, because then there is a higher probability to invoke more other Web services. And also we suggest optimization techniques to get optimized composition result. Optimization processes consist of two phases: one is to remove unnecessary Web services and the other is to find the best starting point of a composition. Test results show that the proposed composing and optimization methods can compose Web services in optimal length in fast way.

Balancing Manual and Automatic Indexing for Retrieval of Paper Abstracts

MEDLINE is a widely used very large database of abstracts of research papers in medical domain. Abstracts in it are manually supplied with keywords from a controlled vocabulary called MeSH. The MeSH keywords assigned to a specific document are subdivided into MeSH major headings, which express the main topic of the document, and MeSH minor headings, which express additional information about the document’s topic. The search engine supplied with MEDLINE uses Boolean retrieval model with only MeSH keywords used for indexing. We show that (1) vector space retrieval model with the full text of the abstracts indexed gives much better results; (2) assigning greater weights to the MeSH keywords than to the terms appearing in the text of the abstracts gives slightly better results, and (3) assigning slightly greater weight to major MeSH terms than to minor MeSH terms further improves the results.

Advanced Clustering Technique for Medical Data Using Semantic Information

MEDLINE is a representative collection of medical documents supplied with original full-text natural-language abstracts as well as with representative keywords (called MeSH-terms) manually selected by the expert annotators from a pre-defined ontology and structured according to their relation to the document. We show how the structured manually assigned semantic descriptions can be combined with the original full-text abstracts to improve quality of clustering the documents into a small number of clusters. As a baseline, we compare our results with clustering using only abstracts or only MeSH-terms. Our experiments show 36% to 47% higher cluster coherence, as well as more refined keywords for the produced clusters.

A New Efficient Clustering Algorithm for Organizing Dynamic Data Collection

We deal with dynamic information organization for more efficient Internet browsing. As the appropriate algorithm for this purpose, we propose modified ART (artificial resonance theory) algorithm, which functions similarly with the dynamic Star-clustering algorithm but performs a more efficient time complexity of O(nk), (k ≪ n) instead of O(n 2 log 2 n) found in the dynamic Star-clustering algorithm. In order to see how fast the proposed algorithm is in producing clusters for organizing information, the algorithm is tested on CLASSIC3 in comparison with the dynamic Star-clustering algorithm.