Optimizing Deep Learning-Based Hate Speech Classification on Social Media Through Frame Semantic Theory and Hyperparameter Tuning

Paper Details

Authors: Mounggam K.; Wisaeng K.

DOI: 10.28924/2291-8639-24-2026-86

Journal: International Journal of Analysis and Applications

Year: 2026

Publisher: Etamaths Publishing

Document Type: Article

Open Access: All Open Access; Gold Open Access

Cited by: 0

Abstract

This study uses deep learning, frame semantic theory, and precise hyperparameter tuning to classify social media hate speech. Three carefully selected hate speech datasets, arranged according to a frame semantic framework, are used to categorize opposing content, helping to better grasp context than simply looking for words. The Synthetic Minority Over-sampling Technique (SMOTE) enhanced minority-class detection to mitigate class imbalance. To find optimal configurations, deep learning architectures, activation functions, and data split algorithms were tested. The best approach used 70% of the data for training and 30% for testing, a model with hidden layers, a Rectifier activation function, and 10 epochs, achieving 96.30% accuracy, 96.95% recall, 99.15% precision, and 0.98 F1-score. These findings show that frame semantic theory, deep learning, and rigorous hyperparameter optimization can significantly improve social media hate speech detection systems. The method lays the foundation for context-aware and socially responsible content control. © 2026 the author(s).

Keywords

deep learning; frame semantics; hate speech classification; hyperparameter tuning

Read Paper Back to Search