Maria Yasir

Academic

Career

Verbatim text

Explainable Ai For Hate Speech Detection For Urdu 
 
In this day and age of social media, information spreads very quickly through blogs, discussion 
forums, videos, posts, images etc. The rate at which the data is generated and consumed has 
brought about numerous issues; Hate Speech being one of them. It is a relatively new concept and 
lacks a universal definition. It covers a wide variety of content such as hate against race, ethnicity, 
gender, sexual orientation, religion, nationality etc. The social media users indulging in such 
practices believe that they are practicing their freedom of speech. However, content used to target 
a certain group of people and defame them or destroy their repute, due to certain personal beliefs, 
cannot be considered as freedom of speech. Hate speech and hate crimes are connected. Majority 
of the present AI models are all black box models. This means that as a user we cannot determine 
why a model gave a certain result. Our Explainable AI models classify the text into 
Hate/Offensive/Neutral speech and alongside provides explanations which are easily understood by 
a layman. A user can enter their own text to check for any hate or offensive speech. Also, a user can 
enter a UserHandle of a twitter profile. It will scrape the most recent tweets, segregate them into 
Urdu, Roman Urdu and English tweets, and provide results on each language separately. The results 
are shown in the form of charts, feature highlights, word clouds and textual explanation. The main 
focus of work is in Urdu language which is an under resourced language. 
 
 
 
 
 
 
 
 
 
 
Technology Used: 
Python, Tensorflow, Keras, Pytorch, Kaggle, 
 Jupyter Notebooks, Shap, Lime, Google Collab 
Supervisor Name: 
Mr. Umair Arshad & Doc. Omer Beg 
Group Members:   
Maria Yasir (I18-0454)

AI enrichment