Update app.py

app.py

@@ -7,8 +7,13 @@ import sentencepiece
 from google.protobuf import text_format
 import re
 
-# Load sentiment analysis pipeline
-sentiment_model = pipeline("sentiment-analysis")
+# Load multilingual social media sentiment analysis model
+sentiment_model = pipeline(
+    "text-classification",
+    model="cardiffnlp/twitter-xlm-roberta-base-sentiment",
+    tokenizer="cardiffnlp/twitter-xlm-roberta-base-sentiment",
+    return_all_scores=False
+)
 
 # Function to analyze text and calculate risk
 def analyze_text(text):
@@ -17,7 +22,8 @@ def analyze_text(text):
     score = result['score']
 
     # Simple risk scoring logic
-    risk_score = score * (1.5 if sentiment == "NEGATIVE" else 1.0)
+    # Adjusted to also flag high-intensity "negative" and "neutral" sentiment
+    risk_score = score * (1.5 if sentiment.upper() in ["NEGATIVE", "LABEL_0"] else 1.0)
     escalation = risk_score > 0.7  # threshold
     return sentiment, risk_score, escalation
 
@@ -30,6 +36,7 @@ def process_input(text):
 title = "SocialAegis MVP"
 description = """
 A sentiment-based escalation engine to detect emotional volatility in social media posts.
+Now using a multilingual, social-media-trained model.
 """
 
 iface = gr.Interface(