Sentence Classification with Deep Learning Method For Virtual Assistant Applications

Abstract

This paper investigates three different deep learning method performance for virtual assistant applications about sentence classification. The classification is based in Turkish texts. For three different model we demonstrate the performance of each model. We investigate Convolutional Neural Network (CNN), Region Convolutional Neural Network (RCNN) and Long Short Term Memory (LSTM) deep learning methods and compare the accuracy results of the related models. Furthermore, we aim to select the best classification model for our dataset.We have researched effect of the hyper parameters to model accuracy and we used best hyper parameters for each methods and we aimed to gain best performance for our dataset.This resarch helps applications like virtual assistant with classification of the sentence and giving the output of the class. The output of classification could be a text, image or document. Benefit of this comparsion of the methods we realized that instance number increses the model accuracy. The best method for our dataset was the Convolutional Neural Networks (CNN) with the %87.3 accuracy.