Semi Supervised Multi Text Classifications for Telugu Documents

Abstract

As the amount of information available on the internet grows at a rapid pace, text classification becomes critical. This data is in an unstructured state and will need to be digitized. Due to the digital nature of these documents, data must be organized by automatically assigning a collection of documents to predefined labels based on their content. To mitigate the growing impact of news text classification, keyword detection approaches based on mostly supervised classification methods have been proposed. However, in practice, the given data is largely unlabeled, necessitating the use of semi-supervised learning techniques instead. We examine the effectiveness of a semi-supervised method for Telugu news articles in this paper. It also addresses some of the most pressing issues in automated text classification, including dealing with unstructured text, handling large numbers of attributes using natural language processing techniques, and dealing with missing metadata due to Telugu's morphological complexity. After classification, semi-supervised clustering is used to classify patterns. Unlabeled texts are used to adapt the centroids, while unlabeled texts are used to capture text cluster silhouette coefficients. To that end, the aim of this study is to use semi-supervised learning methods to investigate the effect of n-gram feature selection on news article text classification. Statistical results classification rate, precision, recall and F-score for news articles are  validated. The results show that the approaches significantly improve the performance.