Combating hate speech using an adaptive ensemble learning model with a case study on COVID-19.

Area of Science:

• Natural Language Processing
• Machine Learning
• Social Media Analysis

Background:

• Social media platforms generate vast amounts of data, making manual hate speech detection inefficient and slow.
• Existing automatic hate speech detection models struggle with cross-dataset generalization and exhibit user bias.
• Hateful content on social media poses a significant challenge due to its subjective nature and rapid dissemination.

Purpose of the Study:

• To develop an adaptive model for automatic hate speech detection that improves cross-dataset generalization.
• To address and overcome the user bias prevalent in annotated social media datasets.
• To enhance the efficiency and accuracy of identifying hateful content across diverse social media contexts.

Main Methods:

• An ensemble learning-based adaptive model was proposed for hate speech detection.
• The model was designed to mitigate user bias inherent in existing datasets.
• Experiments were conducted using various setups, including recent events like COVID-19 and US presidential elections.

Main Results:

• The proposed model demonstrated superior cross-dataset generalization compared to existing methods.
• The model achieved the least performance loss during cross-dataset evaluation.
• No performance drop was observed when limiting the number of tweets per user.

Conclusions:

• The ensemble learning-based adaptive model offers a robust solution for automatic hate speech detection.
• The model effectively generalizes across different datasets and mitigates user bias.
• This approach shows promise for real-world applications in content moderation and online safety.