HATE SPEECH DETECTION FOR TIGRIGNA LANGUAGE ON SOCIAL MEDIA USING DEEP LEARNING

Abstract

Hate speech on social media poses a significant challenge to online safety and social harmony, with far-reaching implications for individuals and communities. Social media platforms are frequently misused for harassment and targeting, with hate being expressed in various forms, such as sexism, racism, and political intolerance. Tigrigna, a widely spoken language in northern Ethiopia and the official language of Eritrea, holds significant cultural and social importance. Despite the growing number of Tigrigna speakers active on social media, research on detecting hate speech in Tigrigna remains scarce. This thesis investigates the design and development of a hate speech detection system for Tigrigna, leveraging a deep learning approach using Bidirectional Long Short-Term Memory (BiLSTM) and Long Short-Term Memory (LSTM) networks. The system was trained on a dataset consisting of 8,583 annotated social media posts, with 2,610 labeled as hate speech and 5,973 labeled as hate-free speech. The performance of the proposed model on the unseen test data gives an accuracy of 92% using the LSTM and 91.5% using Bi-LSTM models. These results suggest that further optimization can be done to enhance the effectiveness of the proposed method on the same dataset. The thesis concludes with recommendations for future research and practical applications.