Hate Speech Prediction October 2023

Developed a robust hate speech detection algorithm aimed at classifying speech as normal or blocked to facilitate content filtration.

Technical stack Used in the Project -

• Utilized the OxAISH-AL-LLM/wiki_toxic dataset from Hugging Face for model training, taking advantage of a pre-existing BERT model. Fine-tuned the last layer and added an additional output layer with two neurons for classification.
• Delivered outstanding results with an exceptional accuracy of 91.95%, a testament to the rigorous training and evaluation processes involved.
• Implemented model deployment through Flask and Docker, ensuring scalability and ease of integration for content filtration solutions.

The Github code is here