app.py

import streamlit as st
import pandas as pd
from transformers import pipeline
import torch  # For GPU checks
import numpy as np
from groq import Client as Groq
import os
from dotenv import load_dotenv
import time
load_dotenv()

os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
# Set page config
st.set_page_config(
    page_title="Restaurant Review Analyzer 🍽️",
    page_icon="🔍",
    layout="wide"
)

# Custom CSS
st.markdown("""
    <style>
    .main {
        padding: 2rem;
    }
    .stProgress > div > div > div {
        background-color: #1f77b4;
    }
    .metric-card {
        background-color: #f8f9fa;
        padding: 1rem;
        border-radius: 0.5rem;
        box-shadow: 0 2px 4px rgba(0,0,0,0.1);
    }
    </style>
""", unsafe_allow_html=True)


def setup_classifier():
    """Initialize the zero-shot classification pipeline with GPU support if available"""
    with st.spinner('Loading classification model... ⚙️'):
        device = 0 if torch.cuda.is_available() else -1  # Use GPU if available
        return pipeline(
            "zero-shot-classification",
            model="joeddav/xlm-roberta-large-xnli",
            device=device
        )


def create_aspect_labels():
    """Create labels for all aspects with positive/negative sentiment"""
    aspects = [
        "food quality",
        "service",
        "ambiance",
        "price",
        "cleanliness",
        "portion size",
        "wait time",
        "menu variety"
    ]

    sentiment_labels = []
    for aspect in aspects:
        sentiment_labels.extend([
            f"positive {aspect}",
            f"negative {aspect}"
        ])

    return aspects, sentiment_labels


def classify_review(classifier, review, sentiment_labels):
    """Classify a single review across all aspects and sentiments"""
    if pd.isna(review) or not isinstance(review, str):
        return {label: 0 for label in sentiment_labels}

    try:
        result = classifier(
            review,
            sentiment_labels,
            multi_label=True
        )
        return dict(zip(result['labels'], result['scores']))
    except Exception as e:
        st.error(f"Error processing review: {e}")
        return {label: 0 for label in sentiment_labels}

def format_summary_for_llm(summary_df):
    """Format the classification summary into a clear text prompt"""
    summary_text = "Restaurant Reviews Analysis Summary:\n\n"

    sentiment_analysis = {}
    for aspect in summary_df.index:
        pos = summary_df.loc[aspect, 'positive_mentions']
        neg = summary_df.loc[aspect, 'negative_mentions']
        total = pos + neg
        if total > 0:
            pos_percent = (pos / total) * 100
            neg_percent = (neg / total) * 100
            difference = pos_percent - neg_percent
            sentiment_analysis[aspect] = {
                'difference': difference,
                'positive_percent': pos_percent,
                'negative_percent': neg_percent,
                'total_mentions': total
            }

    for aspect, metrics in sentiment_analysis.items():
        summary_text += f"{aspect}:\n"
        summary_text += f"- Total Mentions: {metrics['total_mentions']}\n"
        summary_text += f"- Positive Mentions: {metrics['positive_percent']:.1f}%\n"
        summary_text += f"- Negative Mentions: {metrics['negative_percent']:.1f}%\n"
        summary_text += f"- Sentiment Difference: {metrics['difference']:.1f}%\n"
        summary_text += "\n"

    return summary_text

def generate_insights(groq_client, summary_text):
    """Generate insights using Groq API"""
    prompt = f"""You are an expert restaurant consultant analyzing customer feedback data.
    Based on the following customer review analysis summary, provide actionable insights
    and recommendations for the restaurant owner. When analyzing the data:

    - If an aspect has a positive difference of 0.5% or more, consider it a strength
    - If an aspect has a negative difference of 0.5% or more, consider it an area for improvement
    - For differences smaller than 0.5%, consider the aspect neutral or mixed
    - Pay special attention to aspects with high total mentions as they represent stronger customer sentiment

    Analysis Data:
    {summary_text}

    Please provide:
    1. Key Strengths: What's working well (aspects with >0.5% positive difference)
    2. Areas for Improvement: What needs attention (aspects with >0.5% negative difference)
    3. Mixed Reception Areas: Aspects with minimal difference (<0.5%) between positive and negative
    4. Actionable Recommendations: Specific steps based on the analysis
    5. Priority Actions: What should be addressed first, considering both sentiment differences and total mention count

    Format your response in clear sections with bullet points where appropriate.
    Add relevant emojis to make the response more engaging.
    """

    try:
        with st.spinner('Generating insights... 🤔'):
            chat_completion = groq_client.chat.completions.create(
                messages=[{"role": "user", "content": prompt}],
                model="mixtral-8x7b-32768",
                temperature=0.7,
                max_tokens=1500,
            )
            return chat_completion.choices[0].message.content
    except Exception as e:
        st.error(f"Error generating insights: {str(e)}")
        return None

def main():
    # Header
    st.title("🍽️ Restaurant Review Analyzer")
    st.markdown("### Transform your customer feedback into actionable insights! 📊")

    # Sidebar
    st.sidebar.header("📝 Configuration")

    # File upload
    uploaded_file = st.sidebar.file_uploader("Upload your CSV file", type=['csv'])

    if uploaded_file is not None:
        # Read CSV
        try:
            df = pd.read_csv(uploaded_file)
            df.columns = df.columns.str.strip()

            # Validate 'Review' column
            if 'Review' not in df.columns:
                st.error("❌ 'Review' column not found in the CSV file!")
                return

            # Show sample of uploaded data
            st.subheader("📋 Sample Reviews")
            st.dataframe(df[['Review']].head(5), use_container_width=True)

            # Process reviews
            if st.button("🚀 Analyze Reviews"):
                # Initialize classifier
                classifier = setup_classifier()
                aspects, sentiment_labels = create_aspect_labels()

                # Process reviews with progress bar
                results = []
                progress_bar = st.progress(0)
                status_text = st.empty()

                for idx, review in enumerate(df['Review'].head(30)):
                    status_text.text(f"Processing review {idx + 1}/30...")
                    scores = classify_review(classifier, review, sentiment_labels)
                    results.append(scores)
                    progress_bar.progress((idx + 1) / 30)

                results_df = pd.DataFrame(results)

                # Analyze results
                summary = pd.DataFrame()
                for aspect in aspects:
                    pos_col = f"positive {aspect}"
                    neg_col = f"negative {aspect}"

                    summary.loc[aspect, 'positive_mentions'] = (results_df[pos_col] > 0.5).sum()
                    summary.loc[aspect, 'negative_mentions'] = (results_df[neg_col] > 0.5).sum()
                    summary.loc[aspect, 'avg_positive_score'] = results_df[pos_col].mean()
                    summary.loc[aspect, 'avg_negative_score'] = results_df[neg_col].mean()

                # Display summary in columns
                st.subheader("📊 Analysis Summary")
                col1, col2 = st.columns(2)

                with col1:
                    st.markdown("#### 📈 Positive Mentions")
                    for aspect in aspects:
                        st.metric(
                            label=aspect.title(),
                            value=f"{summary.loc[aspect, 'positive_mentions']} reviews",
                            delta=f"{summary.loc[aspect, 'avg_positive_score']:.2%} avg. confidence"
                        )

                with col2:
                    st.markdown("#### 📉 Negative Mentions")
                    for aspect in aspects:
                        st.metric(
                            label=aspect.title(),
                            value=f"{summary.loc[aspect, 'negative_mentions']} reviews",
                            delta=f"{summary.loc[aspect, 'avg_negative_score']:.2%} avg. confidence",
                            delta_color="inverse"
                        )

                # Generate insights
                groq_client = Groq(api_key=os.getenv("GROQ_API_KEY"))

                summary_text = format_summary_for_llm(summary)
                insights = generate_insights(groq_client, summary_text)

                if insights:
                    st.subheader("💡 Key Insights and Recommendations")
                    st.markdown(insights)

        except Exception as e:
            st.error(f"Error processing file: {str(e)}")

    else:
        # Show welcome message and instructions
        st.markdown("""
        ### 👋 Welcome to the Restaurant Review Analyzer!

        To get started:
        1. 📁 Upload your CSV file containing customer reviews
        2. 🔍 Make sure your file has a 'Review' column
        3. 🚀 Click 'Analyze Reviews' to process the data
        4. 📊 Get detailed insights and recommendations

        The analyzer will process the reviews to provide quick insights!
        """)


if __name__ == "__main__":
    main()