Text Document clustering using partial Fractionation and Bisecting K-means

In the era of big data, text clustering has come up as an important technique in fields such as web documents collection, news recommendation systems, health data collection, business data management, health data management, etc. In that context, this paper presents an automatic text document clustering technique to identify distinct groups in large number of unordered documents. The text documents of one group are very much similar to each other and dissimilar to documents in other groups. The grouping criterion is based on a function whose value is either maximized or minimized. In this paper, a new method for the calculation of initial set of clusters is proposed using fractionation method, and a new similarity measure is also proposed using the combination of cosine, link, neighbours, and Gaussian function. Using these functions, both local and global similarity is considered. This work describes an approximate algorithm based on clustering which will try to overcome the drawback of k-means of uncertain number of iterations by fixing the number of iterations, without losing the precision. The proposed approach uses the neighbours and links along with Gaussian function in different characteristics of the clustering algorithm for finding initial clusters in text document clustering. The experimental results performed using sports data documents show that performance of the proposed technique is very promising.