A Hate Speech Detection Model Using Transformer Neural & OpenScript.Js

Presentation title

OpenScript.Js – Empowering Web Design and Development

Abstract

OpenScript.Js is a web frontend framework that simplifies the process of creating beautiful and interactive user interfaces for websites and web applications. Unlike other frameworks, OpenScript.Js embraces the familiar JavaScript language used by web developers every day, making it accessible and easy to use.

With OpenScript.Js, web designers and developers can build stunning user interfaces without the complexity of bundling or modules. It offers a range of powerful features that enable the creation of dynamic and responsive web UIs, enhancing the overall user experience. The framework’s key components, such as UI Components, Markup Generator, File Autoloader, State, Context, Event Emitter, and Router, work together seamlessly to provide a comprehensive toolkit for developers. OpenScript.Js empowers developers to create reusable and modular UI components, manage application state efficiently, handle user interactions, and navigate between different views in a single-page application.

OpenScript.Js brings numerous benefits to both designers and developers. It allows designers to express their creativity by providing an intuitive and flexible framework for building visually appealing interfaces. Developers can leverage OpenScript.Js’s simplicity and elegance to enhance their productivity, as they can focus more on crafting exceptional user experiences rather than dealing with complex setup processes.

Presentation title

A Hate Speech Detection Model Using Transformer Neural Network: A Case of Amharic Language

Abstract

Social media is changing how people communicate, consume information, and collaborate. Among its wide range of functionalities, its usage for spreading Hate speech has become common. For the past few years, social media has been the primary tool for disseminating hate speech in Ethiopia. Consequently, the country has encountered several instabilities.

With the growing awareness and effort to combat hate speech using Natural Language Processing, there have been researches and trials to develop hate speech detection models using the local dialect, Amharic. However, there is still little research done in the field of NLP using the Amharic language. Amharic is one of Ethiopia’s widely spoken dialects with its unique script. This research aimed to develop a hate speech detection model using Transformer Neural Network for the Amharic language. Amharic dataset collected from Facebook posts and comments were used to train and test the model. In the research, fine-tuned

Transformer models such as mBERT, RoBERTa, and mT5 were used to classify Amharic texts as hate speech or not. At the end of the research, the developed models were noticed fast to train. The members and RoBERTa models performed well with a respective accuracy of 91% and 90% while the mT5 model had less performance with an accuracy of 51%. Eventually, these models were used in a web application to detect and report the trends of hate texts on Twitter using the Amharic corpus dataset collected from Twitter. The final models were pushed to the Hugging Face repository and made available to the community.