Detection of Online Hate Comments Based on Feature Embedding and Deep Learning
Salehah Hamzah, Masnizah Mohd, Lailatul Qadri Zakaria · 2023
Nowadays, anti-social behavior is common on social media. One of these behaviors is d disseminating hate-based postings. Authors of hate speech have targeted specific groups of people based on their identities, such as race, religion, country, ethnicity, and gender. They are more commonly referred to as Anti-black, Anti-Muslim, Anti-Asian, xenophobic, misogynistic, and homophobic. The hate-biased content in social media is noisy, uses emojis and contains grammatical errors, making the word meaning using statistical feature extraction and machine learning difficult. Due to the success of pre-trained models in computer vision, this paper presented the utilization of Global Vector (GLoVe), feature embedding model to capture the global context, specifically for race and religion class from social media data and modeling using two deep learning approaches named Convolutional Neural Network (CNN) and Bidirectional Long Short Term Memory (BiLSTM). The results demonstrate that CNN classifier outperforms BiLSTM in terms of F1 score using employing GloVe 100 and Glove 300 dimensions.