Update app.py

app.py

@@ -1,47 +1,61 @@
 import streamlit as st
-import torch
-from transformers import pipeline, AutoModelForSequenceClassification, AutoTokenizer
+from transformers import pipeline
 import numpy as np  
 
-# Initialize the raw model and tokenizer (assuming a custom 5-class model)
-model = AutoModelForSequenceClassification.from_pretrained("EmmaL1/CustomModel_amazon")
-tokenizer = AutoTokenizer.from_pretrained("EmmaL1/CustomModel_amazon")
+# Initialize sentiment analysis pipeline
+sentiment_pipeline = pipeline(model="EmmaL1/CustomModel_amazon")
 
 # Initialize question-answering pipeline
 qa_pipeline = pipeline("question-answering", model="distilbert/distilbert-base-cased-distilled-squad")
 
-# Text classification function
+def calculate_rating(sentiment, confidence):
+    """
+    Improved rating calculation that:
+    1. Uses non-linear mapping for more realistic star distributions
+    2. Handles edge cases better
+    3. Provides more 5-star ratings for highly positive sentiment
+    """
+    if sentiment == "POSITIVE":
+        # Non-linear mapping that gives more weight to higher confidence
+        # This creates a more natural distribution of star ratings
+        if confidence > 0.9:
+            return 5
+        elif confidence > 0.7:
+            return 4
+        elif confidence > 0.5:
+            return 3
+        elif confidence > 0.3:
+            return 2
+        else:
+            return 1
+    else:  # NEGATIVE sentiment
+        # Inverted scale for negative sentiment
+        if confidence > 0.9:
+            return 1
+        elif confidence > 0.7:
+            return 2
+        elif confidence > 0.5:
+            return 3
+        elif confidence > 0.3:
+            return 4
+        else:
+            return 5
+
 def textclassification():
     st.title("Amazon Customer Sentiment Analysis: Ratings & Reasons")
     st.write("Enter a sentence to analyze its rating and reason:")
 
     user_input = st.text_input("Input your text:")
     if user_input:
-        # Tokenize input and get model predictions
-        inputs = tokenizer(user_input, return_tensors="pt", truncation=True, padding=True)
-        with torch.no_grad():  # Line 23
-            outputs = model(**inputs)  # Line 24
-            predictions = outputs.logits  # Line 25
-            predictions = torch.softmax(predictions, dim=-1)  # Line 26
-            predictions = predictions.cpu().detach().numpy()[0]  # Line 27
-
-        # Calculate rating (1-5 stars)
-        rating = np.argmax(predictions) + 1  # Index 0-4 -> 1-5 stars
-
-        # Display the rating
-        st.write(f"The rating is {rating} stars")
-        st.write(f"Prediction probabilities: {predictions}")  # For debugging
-
-        # Question Answering
-        qa_input = {
-            'question': f'Why is the rating {rating} stars?',
-            'context': user_input  
-        }
-        qa_result = qa_pipeline(qa_input)
-        st.write(f"Reasons: {qa_result['answer']}")  
-
-def main():
-    textclassification()
-
-if __name__ == "__main__":
-    main()
+        # Sentiment Analysis
+        sentiment_result = sentiment_pipeline(user_input)
+        sentiment = sentiment_result[0]["label"]
+        confidence = sentiment_result[0]["score"]
+
+        st.write(f"Sentiment: {sentiment}")
+        st.write(f"Confidence Score: {confidence:.2f}")
+
+        # Calculate rating using improved function
+        rating = calculate_rating(sentiment, confidence)
+
+        # Display the ra