ABSTRACT

Recently the text mining has emerged as one of the most important fields of data mining because of most of the searching in the web is done on the basis of provided text, also the increasing use of social web network uses the text as major component and extracting the effective information directly or indirectly requires an efficient grouping algorithm which should be capable of providing efficient clustering. The most widely used techniques use vector space model to find equivalent vector of the text for clustering. The vector space model represents the text on the form of n-tuples numeric array (vector) where each dimension represents the unique word and the value is the weight of that word on the basis of term frequency-inverse document frequency (tf-idf), the problem of the technique is that the unique words count in any document may be very large which will create the similarly long vectors whose processing will require large memory with processing power secondly analysis may be required a bias categorical grouping which not addressed in the above technique. Hence in this paper an efficient clustering approach is presented which uses one dimension for the group of the words representing the similar area of interest with that we have also considered the uneven weighting of each dimension depending upon the categorical bias during clustering. After creating the vector the clustering is performed using seeds-affinity clustering technique. Finally to study the performance of the presented algorithm, it is applied to the benchmark data set Reuters-21578 and compared it for F-measure, Entropy and Execution time with k-means algorithm and the original AP (affinity propagation) algorithm the results shows that the presented algorithm outperforms the others by acceptable margin.