distilbert_finetuned_hatespeech

Dataset Collection:

• The hatespeech dataset is collected from Kaggle, containing 25k Twits.
• The dataset has the three classes 0-Hatespeech 1-Offensive Language 2-Neither https://www.kaggle.com/datasets/mrmorj/hate-speech-and-offensive-language-dataset

Dataset Preparation:

• Change all text to lowercase
• Eliminate URLs
• Use TextBlob for spelling correction
• Convert emojis to text and keep sentiment
• Keep hashtags
• Eliminate @mentions
• Balancing

distilbert-base-uncased model fine-tuned for Hate Speech Detection

• The model is fine-tuned on the dataset, used PyTorch for tokenization
• This model can be used to create the labels for academic purposes or for industrial purposes.
• This model can be used for the inference purpose as well.

Data Fields: 24.783 Label: 0-Hatespeech 1-Offensive Language 2-Neither

Application:

• The model is useful for tweet hatespeech detection
• The model can be used to infer labels, classifying the tweet between the three classes
• You can fine-tune this model for your particular use cases

Model Implementation

• Best Hyperparameters: {'batch_size': 32, 'epochs': 7, 'learning_rate': 1e-05}
  • Accuracy: 0.97
• Model Results:
  • Accuracy: 0.84
  • Precision: 0.83
  • Recall: 0.84
  • F1 Score: 0.83
• Steps to Implement Model: