BanglaBait: using transformers, neural networks & statistical classifiers to detect clickbaits in New Bangla Clickbait Dataset

Abstract

The art of luring us to click on certain content by exploiting our curiosity is recognized as clickbait. Clickbait might be aggravating at times because it is misleading. Several studies have worked on the detection of clickbait in online platforms as we transition from the Information Age to the Age of AI. Nonetheless, predicting clickbait in Bengali new articles is still a work in progress. Here, we use deep learning, the process of extracting pattern or feature from data using neural networks, to determine whether an online Bengali article is clickbait or not. We scrape data from online Bengali news articles, manually annotate them and employ deep nerural network architectures like CNN, Bi-LSTM,Bi-GRU and pre-trained fine-tuning language representation approaches –i.e. BERT, BanglaBERT, M-BERT to provide inputs for various types of classifiers. Finally, we evaluate the classifiers’ outputs and choose the best outcome to predict clickbait in Bengali news articles.