Update app.py

app.py

@@ -305,12 +305,36 @@ def detect_negation(text):
             return True
     return False
 
+def detect_crisis_keywords(text):
+    """
+    Detect crisis/emergency keywords that indicate strong negative emotion
+    """
+    crisis_keywords = [
+        'बचाओ', 'मदद', 'help', 'save',  # Distress calls
+        'मार', 'पीट', 'हिंसा', 'beat', 'hit', 'violence',  # Violence
+        'डर', 'खतरा', 'fear', 'danger',  # Fear/danger
+        'मर', 'मौत', 'death', 'die',  # Death
+        'छोड़', 'leave me', 'stop'  # Desperate pleas
+    ]
+    
+    text_lower = text.lower()
+    for keyword in crisis_keywords:
+        if keyword in text_lower:
+            return True
+    return False
+
 def detect_mixed_emotions(text, prosodic_features):
     """
     Advanced mixed emotion detection using text and audio features
+    CRITICAL: Don't mark crisis/distress as mixed emotions
     """
     text_lower = text.lower()
     
+    # FIRST: Check if this is a crisis situation (never mixed)
+    if detect_crisis_keywords(text):
+        print("⚠️ Crisis keywords detected - NOT treating as mixed emotion")
+        return False
+    
     mixed_indicators = [
         'कभी', 'कभी कभी', 'sometimes',
         'लेकिन', 'पर', 'मगर', 'but', 'however',
@@ -319,24 +343,27 @@ def detect_mixed_emotions(text, prosodic_features):
         'शायद', 'maybe', 'perhaps'
     ]
     
-    positive_words = ['खुश', 'प्यार', 'अच्छा', 'बढ़िया', 'मज़ा', 'happy', 'love', 'good', 'nice']
-    negative_words = ['दुख', 'रो', 'गुस्सा', 'बुरा', 'परेशान', 'sad', 'cry', 'angry', 'bad', 'upset']
+    positive_words = ['खुश', 'प्यार', 'अच्छा', 'बढ़िया', 'मज़ा', 'happy', 'love', 'good', 'nice', 'सुंदर', 'प्रसन्न']
+    negative_words = ['दुख', 'रो', 'गुस्सा', 'बुरा', 'परेशान', 'sad', 'cry', 'angry', 'bad', 'upset', 'निराश', 'चिंता']
     
     has_mixed_indicators = any(ind in text_lower for ind in mixed_indicators)
     has_positive = any(word in text_lower for word in positive_words)
     has_negative = any(word in text_lower for word in negative_words)
     
-    high_pitch_variation = prosodic_features['pitch_std'] > 30
-    high_energy_variation = prosodic_features['energy_std'] > 0.05
+    # Only prosodic if both high pitch AND high energy variation
+    high_pitch_variation = prosodic_features['pitch_std'] > 35
+    high_energy_variation = prosodic_features['energy_std'] > 0.08
     
-    text_mixed = has_mixed_indicators or (has_positive and has_negative)
-    audio_mixed = high_pitch_variation and high_energy_variation
+    # Text must have BOTH opposing emotions to be truly mixed
+    text_mixed = has_mixed_indicators and (has_positive and has_negative)
+    audio_mixed = high_pitch_variation and high_energy_variation and (has_positive and has_negative)
     
     return text_mixed or audio_mixed
 
 def enhanced_sentiment_analysis(text, prosodic_features, raw_results):
     """
     Enhanced sentiment analysis combining text and prosodic features
+    CRITICAL: Properly handle crisis/distress situations
     """
     sentiment_scores = {}
     
@@ -365,34 +392,44 @@ def enhanced_sentiment_analysis(text, prosodic_features, raw_results):
     
     initial_confidence = max(sentiment_scores.values())
     
-    # Negation detection
-    has_negation = detect_negation(text)
-    if has_negation:
-        print("🔄 Negation detected - adjusting sentiment")
-        temp = sentiment_scores['Positive']
-        sentiment_scores['Positive'] = sentiment_scores['Negative']
-        sentiment_scores['Negative'] = temp
-    
-    # Mixed emotions
-    is_mixed = detect_mixed_emotions(text, prosodic_features)
-    if is_mixed:
-        print("🔄 Mixed emotions detected - boosting neutral")
-        neutral_boost = 0.25
-        sentiment_scores['Neutral'] = min(0.7, sentiment_scores['Neutral'] + neutral_boost)
-        sentiment_scores['Positive'] = max(0.1, sentiment_scores['Positive'] - neutral_boost/2)
-        sentiment_scores['Negative'] = max(0.1, sentiment_scores['Negative'] - neutral_boost/2)
-    
-    # Prosodic adjustments
-    if prosodic_features['pitch_std'] > 40 and prosodic_features['energy_mean'] > 0.1:
-        print("🎵 Strong emotional prosody detected")
-        if sentiment_scores['Positive'] > sentiment_scores['Negative']:
-            sentiment_scores['Positive'] = min(0.9, sentiment_scores['Positive'] * 1.15)
-        else:
-            sentiment_scores['Negative'] = min(0.9, sentiment_scores['Negative'] * 1.15)
-        sentiment_scores['Neutral'] = max(0.05, sentiment_scores['Neutral'] * 0.85)
-    elif prosodic_features['energy_mean'] < 0.03 and prosodic_features['pitch_std'] < 15:
-        print("🎵 Calm/neutral prosody detected")
-        sentiment_scores['Neutral'] = min(0.8, sentiment_scores['Neutral'] * 1.2)
+    # CRITICAL: Check for crisis keywords first
+    is_crisis = detect_crisis_keywords(text)
+    if is_crisis:
+        print("🚨 CRISIS DETECTED - Strongly amplifying negative sentiment")
+        # Heavily boost negative sentiment for crisis situations
+        sentiment_scores['Negative'] = min(0.95, sentiment_scores['Negative'] * 1.8)
+        sentiment_scores['Neutral'] = max(0.02, sentiment_scores['Neutral'] * 0.2)
+        sentiment_scores['Positive'] = max(0.01, sentiment_scores['Positive'] * 0.1)
+        is_mixed = False  # Crisis is NEVER mixed emotion
+    else:
+        # Negation detection (only for non-crisis)
+        has_negation = detect_negation(text)
+        if has_negation:
+            print("🔄 Negation detected - adjusting sentiment")
+            temp = sentiment_scores['Positive']
+            sentiment_scores['Positive'] = sentiment_scores['Negative']
+            sentiment_scores['Negative'] = temp
+        
+        # Mixed emotions (only for non-crisis)
+        is_mixed = detect_mixed_emotions(text, prosodic_features)
+        if is_mixed:
+            print("🔄 Mixed emotions detected - boosting neutral")
+            neutral_boost = 0.20  # Reduced from 0.25
+            sentiment_scores['Neutral'] = min(0.65, sentiment_scores['Neutral'] + neutral_boost)
+            sentiment_scores['Positive'] = max(0.1, sentiment_scores['Positive'] - neutral_boost/2)
+            sentiment_scores['Negative'] = max(0.1, sentiment_scores['Negative'] - neutral_boost/2)
+        
+        # Prosodic adjustments (only for non-crisis)
+        if prosodic_features['pitch_std'] > 45 and prosodic_features['energy_mean'] > 0.12:
+            print("🎵 Strong emotional prosody detected")
+            if sentiment_scores['Positive'] > sentiment_scores['Negative']:
+                sentiment_scores['Positive'] = min(0.9, sentiment_scores['Positive'] * 1.2)
+            else:
+                sentiment_scores['Negative'] = min(0.9, sentiment_scores['Negative'] * 1.2)
+            sentiment_scores['Neutral'] = max(0.05, sentiment_scores['Neutral'] * 0.8)
+        elif prosodic_features['energy_mean'] < 0.03 and prosodic_features['pitch_std'] < 15:
+            print("🎵 Calm/neutral prosody detected")
+            sentiment_scores['Neutral'] = min(0.8, sentiment_scores['Neutral'] * 1.2)
     
     # Normalize
     total = sum(sentiment_scores.values())