Update app.py

app.py

@@ -1,43 +1,71 @@
 import gradio as gr
-from detoxify import Detoxify
 import pandas as pd
+from detoxify import Detoxify
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+import torch
+import numpy as np
+import io
 
-# Load model once
-model = Detoxify('original')
+# Load Detoxify multilingual model
+tox_model = Detoxify('multilingual')
 
-# Threshold for flagging a comment as risky
+# Load AI detector model
+ai_tokenizer = AutoTokenizer.from_pretrained("openai-community/roberta-base-openai-detector")
+ai_model = AutoModelForSequenceClassification.from_pretrained("openai-community/roberta-base-openai-detector")
+
+# Thresholds
 TOXICITY_THRESHOLD = 0.7
+AI_THRESHOLD = 0.5  # If >0.5, it's likely AI-generated
+
+def detect_ai_generated(text):
+    with torch.no_grad():
+        inputs = ai_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+        logits = ai_model(**inputs).logits
+        probs = torch.sigmoid(logits).squeeze().item()
+    return round(probs, 4)
 
-def classify_multiple(comments):
-    # Split input into lines
-    comment_list = [c.strip() for c in comments.split('\n') if c.strip()]
-    if not comment_list:
-        return "Please enter at least one valid comment."
+def process_input(file):
+    df = pd.read_csv(file.name)
+    if 'comment' not in df.columns:
+        return "CSV must contain a 'comment' column."
+    
+    comments = df['comment'].astype(str).tolist()
+    tox_results = tox_model.predict(comments)
+    tox_df = pd.DataFrame(tox_results, index=comments).round(4)
 
-    # Predict toxicity scores
-    results = model.predict(comment_list)
-    df = pd.DataFrame(results, index=comment_list).round(4)
+    # Format columns
+    tox_df.columns = [col.replace("_", " ").title().replace(" ", "_") for col in tox_df.columns]
+    tox_df.columns = [col.replace("_", " ") for col in tox_df.columns]
 
-    # Capitalize headers
-    df.columns = [col.replace("_", " ").title().replace(" ", "_") for col in df.columns]
-    df.columns = [col.replace("_", " ") for col in df.columns]
+    # Add warnings
+    tox_df["⚠️ Warning"] = tox_df.apply(lambda row: "⚠️ High Risk" if any(score > TOXICITY_THRESHOLD for score in row) else "✅ Safe", axis=1)
 
-    # Add warning column
-    def check_warning(row):
-        return "⚠️ High Risk" if any(score > TOXICITY_THRESHOLD for score in row) else "✅ Safe"
+    # Add AI detection
+    tox_df["🧪 AI Probability"] = [detect_ai_generated(c) for c in tox_df.index]
+    tox_df["🧪 AI Detection"] = tox_df["🧪 AI Probability"].apply(lambda x: "🤖 Likely AI" if x > AI_THRESHOLD else "🧍 Human")
 
-    df["⚠️ Warning"] = df.apply(check_warning, axis=1)
+    # Store downloadable CSV
+    csv_data = tox_df.copy()
+    csv_data.insert(0, "Comment", tox_df.index)
+    csv_bytes = csv_data.to_csv(index=False).encode()
+    return tox_df, ("toxicity_report.csv", csv_bytes)
 
-    return df
+# Gradio UI
+upload = gr.File(label="📥 Upload .CSV (Must contain 'comment' column)")
+output_table = gr.Dataframe(label="📊 Predictions (Multilingual + AI Detection)")
+download = gr.File(label="📤 Download Predictions")
 
-# UI setup
-iface = gr.Interface(
-    fn=classify_multiple,
-    inputs=gr.Textbox(lines=8, placeholder="Enter one or more comments, each on a new line..."),
-    outputs=gr.Dataframe(label="Toxicity Predictions with Warnings"),
-    title="💬 Toxic Comment Classifier (Multi-Comment, with Warnings)",
-    description="Paste one or more comments. Each comment is scored for toxicity, and flagged as ⚠️ if any label exceeds 0.7."
+app = gr.Interface(
+    fn=process_input,
+    inputs=upload,
+    outputs=[output_table, download],
+    title="🌍 Toxic Comment Classifier + AI Text Detector",
+    description="""
+📥 Upload a .csv file with a 'comment' column.  
+🔍 Each comment will be scored for toxicity (Multilingual model) and AI-generation probability (RoBERTa-based).  
+📤 Download the full report as .csv.
+"""
 )
 
 if __name__ == "__main__":
-    iface.launch()
+    app.launch()