Table based K Nearest Neighbor for Text Classification

Abstract

This article proposes the modified KNN (K Nearest Neighbor)algorithm which receives a table as its input data and is applied tothe text categorization. The motivations of this research are thesuccessful results from applying the table based algorithms to thetext categorizations in previous works and the expectation ofsynergy effect between the text categorization and the wordcategorization. In this research, we define the similarity metricbetween two tables representing texts, modify the KNN algorithm byreplacing the exiting similarity metric by the proposed one, andapply it to the text categorization. The proposed KNN is empiricallyvalidated as the better approach in categorizing texts in newsarticles and opinions. In using the table based KNN algorithm, it iseasier to trace results from categorizing texts.