EiAP-BC: A Novel Emoji Aware Inter-Attention Pair Model for Contextual Spam Comment Detection Based on Posting Text

Detecting spam comments on social media remains a continuously discussed research topic to this day, especially on public figure/celebrity accounts in Indonesia. However, the previous studies only focused on the comments themselves, without considering the context of the posting and the use of emojis in social media. This study proposes a new deep learning model called EiAP-BC (Emoji-aware Inter-Attention Pair BiLSTM CNN) for spam comment detection through a novel approach that considers the contextual information of the posts, enabling the detection of spam comments using the relatedness between the comment and its corresponding post that is usually discarded. This model can also handle emoji content in comments and posts, which is widely used in social media. The model was tested using the SPAMID-PAIR dataset created from social media in the Indonesian language, achieving the highest accuracy of 88% and performing competitively with existing deep learning models. To assess its generalization capabilities, the EiAP-BC model was also evaluated using similar public datasets and models in sentence-pair classification tasks, and an ablation study was conducted to determine the importance of each layer and its coordination. The EiAP-BC model exhibits several advantages in size, training speed, and parameter count compared to existing state-of-the-art models.