Update app.py

app.py

@@ -9,23 +9,34 @@ from sklearn.pipeline import make_pipeline
 sentiment_model = pipeline("text-classification", model="cardiffnlp/twitter-roberta-base-sentiment", top_k=None)
 
 # Define possible sentiment classes
-labels = ["very negative", "negative", "slightly negative", "neutral", "slightly positive", "positive", "very positive", "anger", "joy", "sadness"]
+label_mapping = {
+    "LABEL_0": "very negative",
+    "LABEL_1": "negative",
+    "LABEL_2": "slightly negative",
+    "LABEL_3": "neutral",
+    "LABEL_4": "slightly positive",
+    "LABEL_5": "positive",
+    "LABEL_6": "very positive",
+    "LABEL_7": "anger",
+    "LABEL_8": "joy",
+    "LABEL_9": "sadness"
+}
 
 # Function to get sentiment prediction
 def analyze_sentiment(text):
     results = sentiment_model(text)
-    scores = {res['label']: res['score'] for res in results[0]}
+    scores = {label_mapping.get(res['label'], res['label']): res['score'] for res in results[0]}
     sorted_scores = sorted(scores.items(), key=lambda x: x[1], reverse=True)
     top_label, top_confidence = sorted_scores[0]
     return f"Sentiment: {top_label} (Confidence: {top_confidence:.2f})"
 
 # Explainability function using LIME
 def explain_prediction(text):
-    explainer = lime.lime_text.LimeTextExplainer(class_names=labels)
+    explainer = lime.lime_text.LimeTextExplainer(class_names=list(label_mapping.values()))
     
     def predictor(texts):
         predictions = [sentiment_model(text)[0] for text in texts]
-        return np.array([[pred[label] if label in pred else 0 for label in labels] for pred in predictions])
+        return np.array([[pred[label] if label in pred else 0 for label in label_mapping.values()] for pred in predictions])
     
     exp = explainer.explain_instance(text, predictor, num_features=6)
     return exp.as_list()