Deep Learning for Native Advertisement Detection in Electronic News: A Comparative Study

Native advertising is a type of commercial hybrid content that has successfully targeted online consumers. Native ads have become popular, especially on online platforms. People often don't realize that they are reading sponsored and paid content making native advertising very effective. Despite these benefits, native advertising has provoked strong negative responses, often accompanied by criticism and avoidance of the ads. This study aims to investigate the performance of different word embedding methods (BERT, GloVe, FastText) when used in combination with different deep learning methods (BiLSTM, CNN, LSTM) in detecting native ads in electronic news. The dataset used in this study was collected from the news portal, detik.com. Among these models, the BERT-BiLSTM model achieved the highest accuracy, f1 score, recall, precision and AUC score, all with the same value of 95%, compared to the other models, showing that the BERT-BiLSTM model outperformed the other models.