Automated Detection And Analysis Of Sentiment And Hate Speech In Himtagne Language Social Media Posts And Comments Using Deep Learning

Abstract

Hate speech on social media has become a critical global concern, particularly in under-resourced languages where automated moderation tools are limited or unavailable. In Ethiopia, while some progress has been made for languages such as Amharic, the Himtagne language remains largely unsupported, allowing harmful online content to spread unchecked. This study proposes an automated hate speech detection system for Himtagne social media text using deep learning techniques. A dataset of 9,258 Himtagne Facebook posts and comments was collected and manually annotated into five classes: non-hate, religious, gender-based, racial, and political hate speech. The study compares frequency-based feature extraction (TF-IDF) with embedding-based representation (FastText), and evaluates two classification models: Support Vector Machine (SVM) and Stacked Bidirectional Long Short-Term Memory (SBi-LSTM).