Classifying news articles using feature similarity K nearest neighbor

Abstract

This research proposes the KNN (K Nearest Neighbor) which computes the similarity between data items considering features or attributes as well as one to one values. The assumption of the independency among attributes is the violation against the reality especially in the text classification where words are used as features of texts. In this research, we define the similarity measure which considers both attributes and attribute values, modify the traditional version of KNN using the similarity measure, and apply it to the task of text classification. As benefits from this research, it provides the more compact representations of texts and the better performance. Therefore, the goal of this research is to implement the text categorization system with its more efficient data representations and better performance.