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