AN ANALYSIS OF VERIFIED TWITTER ACCOUNTS: INVESTIGATING THEIR ROLE IN PROPAGATING HATE SPEECH, VIOLENCE, DISCRIMINATION, AND CONFLICT

AbstractIn the digital age, social media has emerged as a central platform for public dis- course, offering unparalleled opportunities for connectivity and community en- gagement. However, this digital landscape is not without its challenges. This the- sis delves into the dynamics of hate speech within Twitter conversations, especially those involving Verified Accounts. Utilizing advanced machine learning models and statistical methods, this study analyzes conversational features to understand the broader conversational context often overlooked in hate speech research. The research employs a variant of BERT (Bidirectional Encoder Representations from Transformers), a transformer-based model adept at capturing contextual relation- ships within texts. BERT’s ability to pre-train deep bidirectional representations makes it highly effective for hate speech and toxic comment classification. Along- side BERT, Locality Sensitive Hashing techniques are used for document classifica- tion, adding another layer of sophistication to the analysis. These state-of-the-art models, fine-tuned for high accuracy, provide a robust framework for classifying toxic content on social media. The thesis also utilizes DistilBERT, a lighter version of BERT that maintains over 95% of the original model’s performance while being more efficient. These models, integrated using the Detoxify Python library, enable a comprehensive analysis of hate speech’s impact on social media conversations. This approach leads to insights that contribute to the development of more effec- tive strategies for mitigating hate speech online, fostering safer and more inclusive digital environments.Keywords: Toxic Tweet, Verified Accounts, Google API, Natural Language Pro- cessing, Machine Learning Models.