Update app.py

app.py

@@ -34,27 +34,46 @@ def full_analysis(text, history):
     if not text:
         return "Entrez une phrase.", "", 0, history
 
-    # 1. Détection sentiment par pipeline
-    result = classifier(text)[0]
-    detected_sentiment = result['label'].lower()
+    # 1. Détecter le sentiment via pipeline (ou Zephyr selon ton architecture)
+    prompt_sentiment = f"""
+You are a financial analyst AI specialized in analyzing news sentiment.
 
-    # 2. Appel Zephyr pour explication basée sur ce sentiment
+Given the following financial news:
+\"{text}\"
+
+Classify the overall sentiment into exactly one of the following categories:
+- positive
+- neutral
+- negative
+
+Only answer with the exact category name, and nothing else.
+Be strict and do not be afraid to classify clearly.
+"""
+    detected_sentiment = call_zephyr_api(prompt_sentiment).lower()
+
+    if detected_sentiment not in ["positive", "neutral", "negative"]:
+        detected_sentiment = "neutral"
+
+    # 2. Générer l'explication avec un prompt explicite
     prompt_explanation = f"""
 You are a financial analyst AI.
 
-Based on the following financial news:
+Given the following financial news:
 \"{text}\"
 
-Explain clearly why the sentiment is {detected_sentiment}.
-Write a concise paragraph.
+Detected sentiment: {detected_sentiment}.
+
+Now, write only your explanation why the sentiment is {detected_sentiment}.
+Do not repeat the instructions or the prompt.
+Respond only with your explanation in a clear and concise paragraph.
 """
     explanation = call_zephyr_api(prompt_explanation)
 
-    # 3. Calcul score de clarté
+    # 3. Calcul du score de clarté
     clarity_score = textstat.flesch_reading_ease(explanation)
-    clarity_score = max(0, min(clarity_score, 100))  # Limité 0-100
+    clarity_score = max(0, min(clarity_score, 100))  # Borné 0-100
 
-    # 4. Stocker historique
+    # 4. Stockage historique
     history.append({
         "Texte": text,
         "Sentiment": detected_sentiment.capitalize(),
@@ -64,6 +83,7 @@ Write a concise paragraph.
 
     return detected_sentiment.capitalize(), explanation, int(clarity_score), history
 
+
 # Fonction pour télécharger historique
 def download_history(history):
     if not history: