Modified K-Means for Document Clustering System

In today’s era of World Wide Web, there is a tremendous proliferation in the amount of digitized text documents. As there is huge collection of documents on the web, there is a need of grouping the set of documents into clusters. Document clustering plays an important role in effectively navigating and organizing the documents. K-Means clustering algorithm is the most commonly document clustering algorithm because it can be easily implemented and is the most efficient one in terms of execution times. The major problem with this algorithm is that it is quite sensitive to selection of initial cluster centroids. The algorithm takes the initial cluster center arbitrarily so it does not always promise good clustering results. If the initial centroids are incorrectly determined, the remaining data points with the same similarity scores may fall into the different clusters instead of the same cluster. To overcome this problem, modified K-Means approach is proposed to improve the quality of clustering in this paper. Unlike the traditional K-Means clustering, the proposed K-Means method can generate the most compact and stable clustering results based on maximum distance initial centroids points instead of random initial centroid points. Moreover, the similar data points are clustered based on maximum probability distribution of data points. Therefore, the proposed method is more effective and converges to more accurate clusters than original K-Means clustering method. In this paper, experimental results are presented in F-measure using 20-News Group standard dataset