Toxicity Mitigation
Rephrasing the offensive comments in Reddit
- Trained a bi-directional GRU model to predict the toxicity score of scraped 100k reddit comments
- Implemented a novel algorithm that leverages POS tags to generate candidate sentences for offensive text and fine-tuned DistilRoBERTa model to fill the masked words of the candidate sentences