Combination of LSTM and CNN for Article-Level Propaganda Detection in News Articles

Abstract

Propaganda is a way of disseminating information, regardless of whether the information is true or not. Propaganda usually uses bias in obscuring the understanding of the propaganda targets. News articles are one of the media that is often used in spreading propaganda. Text classification in the form of propaganda detection in news articles is a crucial thing to do in relation to preventing the spread of the propaganda. Long Short-Term Memory (LSTM) is a variant of the Recurrent Neural Network (RNN) which has been widely used in text classification. However, LSTM has a weakness in the form of a tendency to high bias in extracting context from information through word order. Convolutional Neural Network (CNN) in text analysis can perform important feature extraction through the use of convolutional layers but is weak when assigned to context extraction. This research tries to compare LSTM, CNN and the combination of the two methods in text classification in the form of propaganda detection in news articles. The combination of each method is proved to improve classification performance and also shorten the required running time.