A comprehensive hate speech detection system implementing multiple machine learning and deep learning approaches to classify text into three categories: Hate Speech, Offensive Language, and Neither. The system compares performance across different models including SVM, Decision Trees, LSTM, Bi-LSTM, and Bi-GRU.
from tensorflow.keras.models import load_model
import pickle
import os
def load_pretrained_models(models_dir='trained-models'):
traditional_models = {}
dl_models = {}
try:
print("Loading traditional models...")
for model_type in [ModelType.SVM, ModelType.DECISION_TREE]:
model_name = model_type.value.lower()
with open(os.path.join(models_dir, f'{model_name}.pkl'), 'rb') as f:
traditional_models[model_type] = pickle.load(f)
print("Loading deep learning models...")
for model_type in [ModelType.LSTM, ModelType.BI_LSTM, ModelType.BI_GRU]:
model_name = model_type.value.lower()
model_path = os.path.join(models_dir, f'{model_name}.h5')
dl_models[model_type] = load_model(model_path)
# Load preprocessor
print("Loading preprocessor...")
with open(os.path.join(models_dir, 'preprocessor.pkl'), 'rb') as f:
preprocessor = pickle.load(f)
return traditional_models, dl_models, preprocessor
except Exception as e:
print(f"Error loading models: {str(e)}")
return None, None, None
# Example usage
traditional_models, dl_models, preprocessor = load_pretrained_models()
# Quick Tweet analysis
analyzer = TweetAnalyzer(preprocessor, traditional_models, dl_models)
result = analyzer.analyze_tweet('your offensive tweet')Note
bi-gru and bi-lstm models are not available in the pre-trained directory because of GitHub file size limitations.