Hybrid Bi-LSTM and CNN for Clickbait News Detection on Indonesia News Media

Clickbait is a special trick used to make readers open a news page repeatedly. In a negative meaning, clickbait is a trap visit deliberately made by the holder of a news media. The goal is to get profit from the number of pages visited. Sometimes what the news topic mentioned in the title, do not described in the content. To know this trap, this study proposes a model that can detect whether there is clickbait or not in a news article. This research uses the supervised learning method and take primary data from one of the most popular news media in Indonesia language. To get the characteristic of clickbait, we conducted simple research on each candidate dataset. We found six most influential characteristic that distinguish news containing clickbait or not. The number of data set each is 3,915 for news titles and also for content. Then put them into one as training dataset. The title is in the first sentence of the news content. In addition, we propose a hybrid deep learning method CNN- BiLSTM. The CNN method extract features of dataset and Bi- LSTM build a prediction model. We compare our model with pure Bi-LSTM. We test the models with several pattern of parameter: window size = 5, Bi-LSTM layers = 2, dropout rate = 0.3, f(Adamax) , and batch size = 32. The F1-score of CNN- BiLSTM better then pure BiLSTM.