Update app.py

app.py

@@ -2,12 +2,14 @@ import streamlit as st
 from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
 import torch
 
+# Set page configuration
 st.set_page_config(
     page_title="Review Assistant",
     page_icon="📝",
     layout="centered"
 )
 
+# Custom page styling with CSS
 st.markdown("""
 <style>
     .main-header {
@@ -46,6 +48,7 @@ st.markdown("""
         font-weight: bold;
         margin-bottom: 0.5rem;
     }
+    /* Updated button styling with icon */
     .stButton>button {
         background-color: #2563EB;
         color: white;
@@ -53,10 +56,17 @@ st.markdown("""
         padding: 0.5rem 2rem;
         border-radius: 6px;
         font-weight: 500;
+        display: inline-flex;
+        align-items: center;
+        justify-content: center;
     }
     .stButton>button:hover {
         background-color: #1D4ED8;
     }
+    /* Button icon styling */
+    .button-icon {
+        margin-right: 8px;
+    }
     .footer {
         text-align: center;
         color: #9CA3AF;
@@ -69,12 +79,21 @@ st.markdown("""
 </style>
 """, unsafe_allow_html=True)
 
+# Main page header
 st.markdown("<h1 class='main-header'>Smart Review Analysis Assistant</h1>", unsafe_allow_html=True)
 st.markdown("<p class='sub-header'>Topic Recognition, Sentiment Analysis, and Auto Reply in One Click</p>", unsafe_allow_html=True)
 
-# ------- Load Pipelines -------
+# Function to load ML pipelines - cached to avoid reloading
 @st.cache_resource
 def load_pipelines():
+    """
+    Load all three machine learning pipelines:
+    1. Topic classifier using zero-shot classification
+    2. Sentiment analysis model
+    3. Reply generator for customer service responses
+    
+    Returns the models and topic labels for use in the app
+    """
     # Topic Classification Model (Zero-shot classification)
     topic_labels = [
         "billing", "account access", "customer service", "loans",
@@ -82,29 +101,33 @@ def load_pipelines():
         "branch service", "transaction delay", "account closure", "information error"
     ]
     
-
-    dtype = torch.float32  
+    # Use float32 for better CPU compatibility
+    dtype = torch.float32
     
+    # Load topic classification model
     topic_classifier = pipeline(
         "zero-shot-classification",
         model="MoritzLaurer/deberta-v3-base-zeroshot-v1",
     )
     
-    # Sentiment Analysis Model
+    # Load sentiment analysis model
     sentiment_classifier = pipeline(
         "sentiment-analysis",
         model="cardiffnlp/twitter-roberta-base-sentiment-latest",
     )
     
-    # Reply Generation Model
+    # Load reply generation model
     model_name = "Leo66277/finetuned-tinyllama-customer-replies"
     tokenizer = AutoTokenizer.from_pretrained(model_name)
     model = AutoModelForCausalLM.from_pretrained(model_name)
     
+    # Function to generate customer service replies
     def generate_reply(text):
+        """Generate a customer service reply based on the input text"""
         prompt_text = f"Please write a short, polite English customer service reply to the following customer comment:\n{text}"
         inputs = tokenizer(prompt_text, return_tensors="pt", truncation=True, max_length=512)
         
+        # Generate response with beam search for better quality
         with torch.no_grad():
             gen_ids = model.generate(
                 inputs.input_ids,
@@ -115,47 +138,65 @@ def load_pipelines():
                 early_stopping=True
             )
             
+        # Clean up generated text
         reply = tokenizer.decode(gen_ids[0][inputs.input_ids.shape[1]:], skip_special_tokens=True).strip()
         reply = reply.strip('"').replace('\n', ' ').replace('  ', ' ')
         return reply
     
     return topic_classifier, sentiment_classifier, generate_reply, topic_labels
 
-
+# Page layout and user input section
 st.markdown("### Enter a review for instant analysis")
-example_review = "The people at the call center are inexperienced and lack proper training. I had to call multiple times to resolve a simple issue."
 
+# Updated example review as requested
+example_review = "BOA states on their website that holds are 2-7 days. I made a deposit, and the receipt states funds would be available in 2 days. Now, 13 days later, I am still waiting on my funds, and BOA can't give me a straight answer."
+
+# Text input area for user reviews
 user_review = st.text_area(
     "Please enter or paste a review below:",
     value=example_review,
     height=120
 )
 
-if st.button("Analyze"):
+# Custom button with icon
+analyze_button = st.markdown("""
+<button class="stButton primaryButton">
+    <span class="button-icon">📊</span> Analyze
+</button>
+""", unsafe_allow_html=True)
+
+# Check if button is clicked (using the regular button for functionality)
+if st.button("Analyze", key="hidden_button", help="Click to analyze the review"):
     if not user_review.strip():
+        # Validation check
         st.warning("Please enter a valid review!")
     else:
+        # Show loading spinner during analysis
         with st.spinner("Analyzing your review..."):
+            # Load models if not already loaded
             if "topic_pipe" not in st.session_state:
                 st.session_state.topic_pipe, st.session_state.sentiment_pipe, st.session_state.reply_generator, st.session_state.topic_labels = load_pipelines()
             
-            # Topic Classification
+            # Perform topic classification
             topic_result = st.session_state.topic_pipe(user_review, st.session_state.topic_labels, multi_label=False)
             topic = topic_result['labels'][0]
             
-            # Sentiment Analysis
+            # Perform sentiment analysis
             sentiment_result = st.session_state.sentiment_pipe(user_review)
             sentiment = sentiment_result[0]['label']
             
-            # Auto Reply Generation
+            # Generate auto-reply
             reply_text = st.session_state.reply_generator(user_review)
 
+        # Display results in a two-column layout
         col1, col2 = st.columns(2)
         with col1:
             st.markdown(f"<div class='result-card topic-card'><p class='result-label'>Topic:</p>{topic}</div>", unsafe_allow_html=True)
         with col2:
             st.markdown(f"<div class='result-card sentiment-card'><p class='result-label'>Sentiment:</p>{sentiment}</div>", unsafe_allow_html=True)
         
+        # Display auto-reply suggestion
         st.markdown(f"<div class='result-card reply-card'><p class='result-label'>Auto-reply Suggestion:</p>{reply_text}</div>", unsafe_allow_html=True)
 
-st.markdown("<div class='footer'>© 2024 Review AI Assistant</div>", unsafe_allow_html=True)
+# Page footer
+st.markdown("<div class='footer'>© 2025 Review AI Assistant</div>", unsafe_allow_html=True)