Table based K Nearest Neighbor for Word Categorization in News Articles
Duke Taeho Jo · 2018
In this research, we propose the KNN version where words are encoded into tables, instead of numerical vectors, as the approach to the word categorization. In the previous works, the better text categorization performances from encoding texts into tables than into numerical vectors are shown and we try to make the mutual reinforcement between word and text categorization by connection them with each other. In this research, words are encoded into tables each of which consists of entries of text identifiers and their weights, the similarity metric between two tables which is based on the ratio of intersection to union, is defined, and the KNN algorithm is modified by adopting the proposed similarity metric. This research shows better results of the proposed KNN version which receives a table as input data than the traditional version, in categorizing words which are from news articles. In future, we connect mutually the word categorization with the text categorization for reinforcing them at same time.