Social Choice Theory Based Domain Specific Hindi Stop Words List Construction and Its Application in Text Mining

In this paper, we have given an attempt to create domain specific Hindi stop words list using statistical and knowledge based techniques from prepared textual corpora of different domains. In order to remove the biased raking nature of each technique, Borda’s rule of vote ranking method has been employed for unbiased stop words list construction. We also propose a novel approach called netting ranked performance evaluation (NRPE) to evaluate prepared stop words lists, in which stop words removal is done in leading and trailing fashion based on ascending and descending order of terms. Further, using combined band net (CBN) performance, we demonstrate the ability of each technique in identifying of candidate stop words followed by selection of features for text mining models. The experimental results show that a technique selects good features for classification/clustering needs not necessarily finds the good stop words. Results also show that the final Borda’s lists gives normalized performance over individual technique. This approach guarantees candidate stop word removal, least information dissipation and text mining model performance enhancement.