app.py

import streamlit as st
import requests

# -----------------------------------------------------------
# SETUP: Hugging Face API and Models
# -----------------------------------------------------------
HUGGINGFACE_API_KEY = "your_huggingface_api_key"  # Replace with your API key from https://huggingface.co/settings/tokens
HEADERS = {"Authorization": f"Bearer {HUGGINGFACE_API_KEY}"}

# Hugging Face Model Endpoints (Using Pretrained Models)
CLASSIFIER_API_URL = "https://api-inference.huggingface.co/models/Hate-speech-CNERG/bert-base-uncased-hatexplain"
GENERATOR_API_URL = "https://api-inference.huggingface.co/models/KAXY/GPT2-against-hate"

# -----------------------------------------------------------
# FUNCTION DEFINITIONS
# -----------------------------------------------------------
def detect_harmful_content(text):
    """Detects harmful content using a Hugging Face model."""
    payload = {"inputs": text}
    response = requests.post(CLASSIFIER_API_URL, headers=HEADERS, json=payload)

    if response.status_code != 200:
        return [{"category": "Error", "score": 0, "message": "Failed to fetch response"}]
    
    results = response.json()

    detected = []
    threshold = 0.5  # Adjust threshold for sensitivity
    for result in results:
        if result.get('score', 0) >= threshold:
            detected.append({"category": result.get('label', 'Unknown'), "score": result.get('score', 0)})
    
    return detected

def generate_mitigation_response(text, detected_categories):
    """Generates a mitigation response based on detected harmful speech."""
    if not detected_categories:
        return "✅ Content appears safe. No harmful content detected."

    categories_str = ", ".join([cat["category"] for cat in detected_categories])
    prompt = (f"The following content has been flagged for {categories_str}:\n\n"
              f"\"{text}\"\n\n"
              "Please generate a respectful and informative moderation response.")

    payload = {"inputs": prompt, "parameters": {"max_length": 150}}
    response = requests.post(GENERATOR_API_URL, headers=HEADERS, json=payload)

    if response.status_code != 200:
        return "⚠️ Error: Could not generate a response."

    generated = response.json()
    return generated[0].get('generated_text', "No response generated.")

# -----------------------------------------------------------
# STREAMLIT USER INTERFACE
# -----------------------------------------------------------
st.set_page_config(page_title="Hate Speech Detector", layout="centered")

st.title("🔍 AI-Powered Hate Speech Detection & Mitigation")
st.markdown("Detects **hate speech, misinformation, and cyberbullying** in social media posts.")

# User Input
user_input = st.text_area("✏️ Enter the text to analyze:", height=150)

if st.button("Analyze"):
    if user_input.strip() == "":
        st.error("⚠️ Please enter some text to analyze.")
    else:
        st.markdown("### 📊 Analysis Results")
        detected = detect_harmful_content(user_input)

        if detected and detected[0].get("category") != "Error":
            for d in detected:
                st.write(f"**Category:** {d['category']} | **Confidence:** {d['score']:.2f}")
        else:
            st.write("✅ No harmful content detected.")

        st.markdown("### 💡 Mitigation Response")
        mitigation_response = generate_mitigation_response(user_input, detected)
        st.write(mitigation_response)