import gradio as gr
import joblib
import os
def load_model():
"""Load model from local files in the Space"""
try:
print("Loading model from local files...")
print("\nFiles in current directory:")
for f in os.listdir("."):
if f.endswith((".jobilib", ".pkl", ".txt")):
print(f" - {f}")
model_files = [
"tfidf_logreg_best.jobilib",
"model.jobilib",
"model.pkl"
]
model = None
for mf in model_files:
if os.path.exists(mf):
print(f"\nLoading model: {mf}")
model = joblib.load(mf)
print(f"✅ Model loaded from {mf}")
break
if model is None:
print("❌ No model file found!")
return None
vectorizer_files = [
"vocab.txt",
"vocab",
"vectorizer.jobilib",
"tfidf.jobilib"
]
vectorizer = None
for vf in vectorizer_files:
if os.path.exists(vf):
print(f"Loading vectorizer: {vf}")
vectorizer = joblib.load(vf)
print(f"✅ Vectorizer loaded from {vf}")
break
if vectorizer is None:
print("⚠️ Vectorizer not found")
# Try to load label encoder
label_encoder = None
if os.path.exists("label_encoder.jobilib"):
print("Loading label encoder...")
label_encoder = joblib.load("label_encoder.jobilib")
print("✅ Label encoder loaded")
return {
"model": model,
"vectorizer": vectorizer,
"label_encoder": label_encoder
}
except Exception as e:
print(f"❌ Error loading model: {str(e)}")
import traceback
print(traceback.format_exc())
return None
# Load model on startup
print("Starting model loading...")
model_components = load_model()
if model_components is None:
print("⚠️ Model loading failed!")
else:
print("✅ All models loaded successfully!")
def predict(text):
"""Predict cyberbullying category"""
if not text.strip():
return "⚠️ Please enter some text.
"
try:
# Check if models are loaded
if model_components is None:
return "❌ Model not loaded. Check logs.
"
model = model_components["model"]
vectorizer = model_components["vectorizer"]
label_encoder = model_components["label_encoder"]
# Check if vectorizer exists
if vectorizer is None:
return "❌ Vectorizer not available
"
# Vectorize the input text
text_vector = vectorizer.transform([text])
# Get prediction
prediction = model.predict(text_vector)[0]
# Get confidence score
try:
probabilities = model.predict_proba(text_vector)[0]
score = max(probabilities)
except:
score = 0.8
# Decode label if encoder exists
if label_encoder is not None:
try:
label = label_encoder.inverse_transform([prediction])[0]
except:
label = str(prediction)
else:
label = str(prediction)
print(f"Prediction: {label}, Score: {score:.4f}")
# Category definitions
cyberbullying_types = {
"age": {"emoji": "👶", "color": "#ff6b6b", "text": "Age-Based Cyberbullying"},
"gender": {"emoji": "⚥️", "color": "#ff8c42", "text": "Gender-Based Cyberbullying"},
"ethnicity": {"emoji": "🌍", "color": "#ffa502", "text": "Ethnicity-Based Cyberbullying"},
"religion": {"emoji": "🙏", "color": "#ff6b9d", "text": "Religion-Based Cyberbullying"},
"other_cyberbullying": {"emoji": "⚠️", "color": "#ff4757", "text": "Other Cyberbullying Detected"},
"not_cyberbullying": {"emoji": "✅", "color": "#00ff64", "text": "Safe Message"}
}
# Get category info
label_lower = str(label).lower().strip()
category = cyberbullying_types.get(
label_lower,
cyberbullying_types.get(label, cyberbullying_types["not_cyberbullying"])
)
# Safe message
if label_lower == "not_cyberbullying":
return f"""
{category['emoji']}
{category['text']}
Confidence: {score:.2%}
"""
else:
# Cyberbullying message
return f"""
{category['emoji']}
{category['text']}
{label}
Confidence: {score:.2%}
"""
except Exception as e:
import traceback
error_msg = traceback.format_exc()
print(f"ERROR in prediction: {str(e)}")
print(error_msg)
return f"❌ Error: {str(e)}
"
with gr.Blocks(css="""
🛡️ Cyberbullying Detection System
")
gr.Markdown("End-to-end NLP system for detecting cyberbullying, including religion-based abuse
")
with gr.Group():
text_input = gr.Textbox(
lines=4,
placeholder="Enter a message to analyze...",
label="Input Text"
)
detect_btn = gr.Button("🔍 Detect", variant="primary")
output = gr.HTML()
detect_btn.click(
fn=predict,
inputs=text_input,
outputs=output
)
if __name__ == "__main__":
demo.launch()