KNN based word categorization considering feature similarity

초록

In this research, we propose that the K Nearest Neighbor should be used for categorizing words semantically, considering the feature similarities. In the reality, the dependencies and relations among features are available; texts as features for encoding words into numerical vectors tend to have their similarities with others. In this research, we define the similarity measure considering both feature values and features and use it for modifying the K Nearest Neighbor as the approach to the word categorization. As the benefits from this research, we obtain the potential possibility of more compact representations of words and the improvement of their discriminations among even sparse vectors. Hence, the goal of this research is to implement the word categorization systems with the benefits. © 2019 ICAI 2015 - WORLDCOMP 2015. All rights reserved.

키워드