Summary
International Symposium on Nonlinear Theory and Its Applications
2015
Session Number:A5L-D
Session:
Number:A5L-D-4
Relation Classification Through Substring Representations Using Nonlinear Classifiers
Zhan Jin, Chihiro Shibata, Kazuya Tago,
pp.389-392
Publication Date:2015/12/1
Online ISSN:2188-5079
Summary:
Semantic relation classification can be considered as a multiclass classification problem. Richer and higher quality feature sets lead to better performance when using traditional features. This tendency also increases the dimensions of the feature space, resulting in an increased processing time, and leads to lower classification accuracy when using nonlinear classifiers. We introduce an approach to build features for relation classification consisting of only low-dimensional vectors representing substrings between two words, called substring vectors. In this paper on substring vectors, we survey the relationship between the numbers of dimensions and the obtained accuracies when nonlinear classifiers are applied. Through experimental results, we found that our approach using relatively low-dimensional representations achieves a sufficiently high accuracy that is better than most existing approaches. Furthermore, we utilize autoencoders for reconstruction and decrease the number of dimensions; finally, we obtain better classification results than before.