alphaCalson/Final-Capstone

This is my Capstone project for hate speech detection using also minority languages in South Africa (NGUNI languages) # Breaking Barriers: Hate Speech Detection in South African Indigenous Languages ## Executive Summary This project aims to develop a comprehensive solution for hate speech detection specifically for South African indigenous languages, particularly the Nguni languages (isiXhosa, isiZulu, etc.). These languages are underrepresented in Natural Language Processing (NLP). The system utilizes three machine learning architectures: Capsule Network, LSTM, and CNN, tailored to capture the nuances of hate speech in online conversations. The project's ultimate goal is to provide a robust tool for making online spaces safer for speakers of these languages, filling a crucial gap in the existing hate speech detection landscape. ## Business Domain This project is designed for use in online content moderation on platforms like social media, news websites, and forums. The intended users are platform administrators, content moderators, and policymakers who need to detect and prevent hate speech across South African indigenous languages. It also supports government initiatives to curb online hate speech and foster inclusive digital spaces. ## Problem Statement Online platforms see an increase in toxic and hateful speech, especially in underrepresented languages like those spoken in South Africa. Current hate speech detection systems are tailored primarily for globally dominant languages (like English), leaving a significant gap in coverage for Nguni languages. This project addresses the challenges of detecting hate speech in these languages and provides a tool for more inclusive moderation. ## Objectives - Develop a multilingual hate speech detection system for Nguni languages using CapsuleNet, LSTM, and CNN architectures. - Achieve high accuracy (90% or higher) in detecting toxic comments in isiNdebele, isiXhosa, isiZulu, Siswati, and English. - Create an intuitive Gradio application to allow platform administrators to utilize the models for hate speech detection. ## Scope # …