Improved Graph Based K-NN Text Classification

This paper presents an improved graph based k-nn algorithm for text classification. Most of the organization are facing problem of large amount of unorganized data. Most of the existing text classification techniques are based on vector space model which ignores the structural information of the document which is the word order or the co-occurrences of the terms or words. In this paper we have used the graph based representation of the text in which structural information of the text document is taken into consideration. Feature selection phase plays a very important role in classification. The emphasis has been on effective feature selection methods using both standard as well globalized methods of feature selection which are MI+Chi, RMI+Chi (standard method) and WT (localized method). The dataset that had been used is self made English text document of five different categories. The final result had shown that it is not always that a standard method of feature selection will improve the categorization but a localized method that is Weight Of Terms [WT] can also improve