|
|
 |
| |
|
|
|
Semantic Safeguards: Harnessing BERT and Advanced Deep Learning Models Outperforming in the Detection of Hate Speech on Social Networks |
|
|
|
PP: 811-825 |
|
|
|
doi:10.18576/amis/180413
|
|
|
|
Author(s) |
|
|
|
Deema Mohammed Alsekait,
Ahmed Younes Shdefat,
Zakwan AlArnaout,
Nermin Rafiq Mohamed,
Hanaa Fathi,
Diaa Salama AbdElminaam,
|
|
|
|
Abstract |
|
|
| This paper presents an innovative approach for hate speech detection on social media platforms utilizing optimized deep
learning algorithms. Capitalizing on the strengths of four machine learning algorithms (Decision Trees, Support Vector Machines,
Naive Bayes, and K-Nearest Neighbors), two deep learning algorithms (Bidirectional Long Short-Term Memory and Recurrent Neural
Networks), and a transformer model (Bidirectional Encoder Representations from Transformers, BERT), this research aims to classify
text as hate speech efficiently. By implementing feature extraction techniques—TF-IDF for machine learning models and embedding
layers for deep learning and transformer models—we leverage two datasets comprising English tweets from Twitter and Facebook. The
results indicate a superior performance of the BERT model, achieving an impressive 95% accuracy on the HSOL dataset and 67% on
the HASOC dataset, thus significantly advancing the hate speech detection methodology. This paper’s methods and findings enhance
the existing body of knowledge and provide a reliable model for improving online social interaction safety. The novelty of our work
lies in the comprehensive preprocessing and the application of BERT in this context, marking a significant scientific contribution with
practical implications for creating a more inclusive online community. |
|
|
|
|
|
|
|
