A Comparative Analysis of SVM, LSTM and CNN-RNN Models for the BBC News Classification

Abstract

AbstractDeveloping technology, easier access to internet networks and the increase in the widespread use of the internet cause more text data to be produced than ever. Analyzing and classifying this data is one of the most difficult problems. Hence, automatic classification of text data is very important. In this study, the word embedding method, machine learning, deep learning and a hybrid model is proposed to classify news texts. GloVe method is used as a word embedding model and SVM (Support Vector Machine), LSTM (Long Short-Term Memory), CNN (Convolutional Neural Network) + RNN (Recurrent Neural Networks) models are used to classify data. Also, various optimization algorithms have been tested in deep learning models and the results of these algorithms are in a comparative manner. BBC-News dataset is used to classify news texts. The best accuracy results have been obtained as 98.0% in the SVM model, 97.0% in the LSTM model and 96.0% in the CNN + RNN model.KeywordsText document classificationNatural language processingBBC news classificationGloVeHybrid model