This repository contains the main deliverables and software designed to produce the results seen in my MEng Final Year Project thesis for titled: "Hate Speech Detection and Hate Network Detection on Social Media".
The current problem with social media is the large reliance on the presence of non-malicious actors. However, with the spreadability of social media ensures that malicious actors will be able to spread hateful views very easily. This project provides practical methodologies to detect hate speech through the development of research and the design of a hate speech detection model. These models were further developed to be utilised in a hateful network detector that has successfully identified hate networks on the Twitter platform. The hateful network detector can be successfully deployed to find chains and communities of hate speech users embedded within non-hate speech communities. The models investigated included K-NNs, transformers and GRUs have been used with various sets of tweets to attempt to produce a model that can easily distinguish hate and non-hate speech instances. The work in this paper also determined the effect of preprocessing on the effectiveness of a model with extensive testing performed to evaluate the effectiveness of the models and the type of preprocessing employed. This research further ponders the question as to why the methods implemented in this paper could not be utilised earlier by the Twitter platform as the methods in this paper showcase that self regulation by the social media platforms should be possible.
I would like to thank Professor Jeremy Pitt and Dr. David Burth Kurka for their continuous advice and supervision of the project in addition to their continuous input and giving me the freedom of choice in how to construct and tailor the project to my interests. I would also like to thank my friends and family for continuous help and support throughout my four years at Imperial and allowing me to grow through my experience at university.
