Convolutional Long Short-Term Memory Implementation for Indonesian News Classification

Abstract

Text classification is now a well-studied field, particularly in Natural Language Processing (NLP). The text classification can be carried out using various methods, one of which is deep learning. Deep learning methods such as RNN, CNN, and LSTM are the most frequent methods used for text classification. This research aims to analyze the implementation of two deep learning methods combination, namely CNN and LSTM (C-LSTM), to classify Indonesian news texts. News texts used as data in this study were collected from Indonesian news portals. The obtained data were then divided into three categories based on their scope: "National," "International," and "Regional." Three research variables were tested in this study: the number of documents, the batch size value, and the learning rate value of the built C-LSTM. The experimental results showed that the F1-score obtained from the classification results using the C-LSTM method was 93.27%. The F1-score value generated by the C-LSTM method was higher than that of CNN (89.85%) and LSTM (90.87%). In summary, the combination method of two deep learning methods, namely CNN and LSTM (C-LSTM), outperforms CNN and LSTM.