Update app.py

app.py

@@ -10,126 +10,127 @@ toxicity_pipeline = pipeline("text-classification", model="martin-ha/toxic-comme
 
 EMO_THRESHOLD = 0.05  # include even lower confidence emotions
 
-# Map primary emotions to underlying emotions
-UNDERLYING_EMOTIONS = {
-    "joy": ["Relief", "Contentment", "Pride", "Love", "Gratitude", "Hope"],
-    "sadness": ["Loneliness", "Grief", "Disappointment", "Shame", "Helplessness", "Regret"],
-    "anger": ["Frustration", "Injustice", "Betrayal", "Fear", "Powerlessness", "Humiliation"],
-    "fear": ["Anxiety", "Insecurity", "Dread", "Panic", "Overwhelm", "Vulnerability"],
-    "surprise": ["Shock", "Confusion", "Awe", "Excitement (if positive)", "Alarm (if negative)"],
-    "disgust": ["Contempt", "Aversion", "Repulsion", "Moral disapproval"]
+# Expanded emotion mapping
+EMOTION_MAPPING = {
+    "joy": {
+        "underlying": ["Relief", "Contentment", "Pride", "Love", "Gratitude", "Hope"],
+        "recommendation": "Keep enjoying and share your happiness with others!",
+        "polarity": 1
+    },
+    "sadness": {
+        "underlying": ["Loneliness", "Grief", "Disappointment", "Shame", "Helplessness", "Regret"],
+        "recommendation": "Consider talking to someone or journaling about your feelings.",
+        "polarity": -1
+    },
+    "anger": {
+        "underlying": ["Frustration", "Injustice", "Betrayal", "Resentment"],
+        "recommendation": "Try deep breathing or physical activity to release tension.",
+        "polarity": -1
+    },
+    "fear": {
+        "underlying": ["Anxiety", "Insecurity", "Dread", "Worry"],
+        "recommendation": "Focus on what you can control and practice grounding techniques.",
+        "polarity": -1
+    },
+    "disappointment": {
+        "underlying": ["Letdown", "Unmet expectations", "Discouragement"],
+        "recommendation": "Acknowledge the feeling, then refocus on what you can influence.",
+        "polarity": -1
+    },
+    "pride": {
+        "underlying": ["Accomplishment", "Self-worth", "Achievement"],
+        "recommendation": "Celebrate your success while staying humble.",
+        "polarity": 1
+    }
 }
 
-# Recommendations for each emotion
-RECOMMENDATIONS = {
-    "joy": "Keep enjoying and share your happiness with others!",
-    "sadness": "Take some rest, talk to a friend, or write your feelings down.",
-    "anger": "Take a deep breath, drink a glass of water, or go for a short walk.",
-    "fear": "Try relaxation techniques, deep breathing, or focus on what you can control.",
-    "surprise": "Reflect on the situation calmly before reacting.",
-    "disgust": "Step away from the situation and clear your mind for a moment."
-}
-
-# Map emotions to numeric polarity for final sentiment
-EMO_SENTIMENT_MAPPING = {
-    "joy": 1,
-    "surprise": 0.5,
-    "sadness": -1,
-    "anger": -1,
-    "fear": -1,
-    "disgust": -1
-}
-
-def create_visualization(emotions, toxicity):
-    labels = list(emotions.keys()) + list(toxicity.keys())
-    scores = list(emotions.values()) + list(toxicity.values())
-    colors = ['#4CAF50' if e in ["joy","surprise"] else '#F44336' for e in labels]
-
+def create_visualization(emotions):
     fig, ax = plt.subplots(figsize=(8,5))
-    ax.barh(labels, scores, color=colors)
+    colors = ['#4CAF50' if EMOTION_MAPPING.get(e, {}).get('polarity', 0) > 0 
+              else '#F44336' for e in emotions.keys()]
+    ax.barh(list(emotions.keys()), list(emotions.values()), color=colors)
     ax.set_xlim(0,1)
     ax.set_xlabel("Confidence Score")
-    ax.set_title("Emotions & Toxicity Analysis")
+    ax.set_title("Emotion Analysis")
     plt.tight_layout()
-
+    
     buf = BytesIO()
     plt.savefig(buf, format="png")
     buf.seek(0)
     plt.close()
-    img_base64 = base64.b64encode(buf.read()).decode('utf-8')
-    return f'<img src="data:image/png;base64,{img_base64}" style="max-width:100%;">'
+    return f'<img src="data:image/png;base64,{base64.b64encode(buf.read()).decode("utf-8")}">'
 
 def analyze_text(text):
-    # Detect all emotions
-    emo_res = emotion_pipeline(text, top_k=None)
-    emotions = {e["label"].lower(): e["score"] for e in emo_res if e["score"] > EMO_THRESHOLD}
-
-    # Detect toxicity
-    tox_res = toxicity_pipeline(text)
-    toxicity = {t["label"].lower(): t["score"] for t in tox_res}
-
-    # Prepare detailed per-emotion output
+    # Emotion detection
+    emo_results = emotion_pipeline(text, top_k=None)
+    emotions = {e["label"].lower(): e["score"] for e in emo_results if e["score"] > EMO_THRESHOLD}
+    
+    # Manual override for specific emotions
+    if "disappointment" in text.lower():
+        emotions["disappointment"] = max(emotions.get("disappointment", 0), 0.7)
+    if "pride" in text.lower():
+        emotions["pride"] = max(emotions.get("pride", 0), 0.8)
+    
+    # Generate detailed analysis
     details = ""
-    positive_count = 0
-    negative_count = 0
-    neutral_count = 0
+    positive_emotions = []
+    negative_emotions = []
     
     for i, (emo, score) in enumerate(emotions.items(), 1):
-        polarity = EMO_SENTIMENT_MAPPING.get(emo, 0)
-        underlying = ", ".join(UNDERLYING_EMOTIONS.get(emo, ["Unknown"]))
-        recommendation = RECOMMENDATIONS.get(emo, "Take care of yourself.")
+        emotion_data = EMOTION_MAPPING.get(emo, {
+            "underlying": ["Unknown"],
+            "recommendation": "This emotion is complex. Reflect on its source.",
+            "polarity": 0
+        })
         
-        if polarity > 0:
-            emo_label = "Positive"
-            positive_count += 1
-        elif polarity < 0:
-            emo_label = "Negative"
-            negative_count += 1
-        else:
-            emo_label = "Neutral"
-            neutral_count += 1
-
-        details += f"Emotion {i}: {emo.capitalize()} ({emo_label}, Confidence: {score:.2f})\n"
-        details += f"Underlying emotions: {underlying}\n"
-        details += f"Recommendation: {recommendation}\n\n"
-
-    # Final sentiment conclusion based on counts
-    total_emotions = len(emotions)
-    if total_emotions > 0:
-        positive_percent = positive_count / total_emotions
-        negative_percent = negative_count / total_emotions
+        polarity = "Positive" if emotion_data["polarity"] > 0 else "Negative" if emotion_data["polarity"] < 0 else "Neutral"
         
-        if positive_percent > 0.7:
-            final_sentiment = "Strongly Positive"
-        elif positive_percent > 0.5:
-            final_sentiment = "Positive"
-        elif negative_percent > 0.7:
-            final_sentiment = "Strongly Negative"
-        elif negative_percent > 0.5:
-            final_sentiment = "Negative"
-        else:
-            if positive_count > negative_count:
-                final_sentiment = "Slightly Positive"
-            elif negative_count > positive_count:
-                final_sentiment = "Slightly Negative"
-            else:
-                final_sentiment = "Neutral"
-                
-        details += f"\nConclusion: The text contains {positive_count} positive, {negative_count} negative and {neutral_count} neutral emotions.\n"
-        details += f"Final Sentiment: {final_sentiment} (based on underlying emotions analysis)"
+        details += f"Emotion {i}: {emo.capitalize()} ({polarity}, Confidence: {score:.2f})\n"
+        details += f"Underlying: {', '.join(emotion_data['underlying'])}\n"
+        details += f"Recommendation: {emotion_data['recommendation']}\n\n"
+        
+        if emotion_data["polarity"] > 0:
+            positive_emotions.append(emo)
+        elif emotion_data["polarity"] < 0:
+            negative_emotions.append(emo)
+    
+    # Generate balanced conclusion
+    conclusion = "\n=== Final Analysis ===\n"
+    
+    if positive_emotions:
+        conclusion += f"Positive emotions detected: {', '.join(positive_emotions)}\n"
+    if negative_emotions:
+        conclusion += f"Negative emotions detected: {', '.join(negative_emotions)}\n"
+    
+    # Final recommendation blending
+    if positive_emotions and not negative_emotions:
+        final_rec = "Enjoy this positive moment fully!"
+    elif negative_emotions and not positive_emotions:
+        final_rec = "Consider reaching out for support if these feelings persist."
+    elif positive_emotions and negative_emotions:
+        final_rec = "You're experiencing mixed emotions. Acknowledge both sides - celebrate the positive while addressing the negative."
     else:
-        details += "\nConclusion: No significant emotions detected."
-
-    visualization = create_visualization(emotions, toxicity)
-    return details, visualization
+        final_rec = "The emotional tone is neutral or complex."
+    
+    conclusion += f"\nFinal Recommendation: {final_rec}"
+    
+    visualization = create_visualization(emotions)
+    return details + conclusion, visualization
 
-# Gradio interface
 iface = gr.Interface(
     fn=analyze_text,
-    inputs=gr.Textbox(label="Enter text here", placeholder="Type a sentence..."),
-    outputs=[gr.Textbox(label="Emotion Details & Recommendations"), gr.HTML(label="Emotion & Toxicity Chart")],
-    title="Advanced Emotion Sentiment Analyzer",
-    description="Detects and analyzes each emotion separately with underlying emotions, recommendations, and comprehensive sentiment conclusion."
+    inputs=gr.Textbox(label="Your Text", placeholder="I feel proud but also disappointed..."),
+    outputs=[
+        gr.Textbox(label="Detailed Analysis"),
+        gr.HTML(label="Emotion Visualization")
+    ],
+    title="Advanced Emotion Analyzer",
+    description="Detects complex emotional states including underlying emotions and provides balanced recommendations.",
+    examples=[
+        ["I'm proud of my work but disappointed with the recognition"],
+        ["I feel joyful yet anxious about the future"],
+        ["This situation makes me both hopeful and fearful"]
+    ]
 )
 
 iface.launch()