GenAI-customer_reviews-processor

Customer Satisfaction Processing Bot

Overview

This tool processes customer satisfaction survey data to:

1. Classify sentiment of customer reviews (positive/negative)
2. Send customized discounts for negative reviews
3. Recommend 3 best trips for positive reviews
4. Extract mentioned locations using Named Entity Recognition (NER)
5. Compare classification metrics against ground truth data
6. Save all output to CSV files

Features

Base Features

• Sentiment analysis of customer reviews
• Personalized responses for both positive and negative reviews
• Trip recommendations for positive reviews
• Discount codes for negative reviews
• CSV output of processed reviews

Extra Features

• Named Entity Recognition (NER) to extract mentioned locations
• Sentiment analysis metrics calculation (accuracy, precision, recall, F1 score)
• Comparison against ground truth sentiment data

Setup

1. Install dependencies:

pip install -r requirements.txt

2. Download spaCy language model:

python -m spacy download en_core_web_sm

3. Set your OpenAI API key:

export OPENAI_API_KEY=your_api_key_here

Usage

Process Reviews

Run the processor:

cd src
python processor.py

This will:

1. Process all reviews in data/customer_surveys.json
2. Save processed reviews to data/processed_reviews.csv
3. Save extracted locations to data/extracted_locations.csv
4. Display sentiment analysis metrics

Generate Visualizations

After processing the reviews, generate visualizations:

cd src
python visualize.py

This will create:

1. confusion_matrix.png: Confusion matrix for sentiment analysis
2. metrics_chart.png: Bar chart of accuracy, precision, recall, and F1 score
3. satisfaction_by_sentiment.png: Average satisfaction scores by sentiment
4. top_locations.png: Top 10 mentioned locations
5. entity_types.png: Distribution of entity types

Output Files

1. processed_reviews.csv: Contains all processed reviews with:

   • Original review text
   • Detected sentiment
   • Reasoning for sentiment classification
   • Ground truth sentiment
   • Customer satisfaction score
   • Personalized response message
   • Recommended trips (for positive reviews)
   • Discount code (for negative reviews)

2. extracted_locations.csv: Contains all extracted locations with:

   • Review ID
   • Location name
   • Entity type

Implementation Details

The system uses:

• LangChain for orchestrating the LLM workflow
• OpenAI's GPT-4o for sentiment analysis and response generation
• spaCy for Named Entity Recognition
• scikit-learn for calculating sentiment analysis metrics
• Matplotlib for visualization of metrics and results

Architecture

1. Data Processing Pipeline:

   • Load customer review data from JSON
   • Perform sentiment analysis with LLM
   • Generate personalized responses based on sentiment
   • Extract locations using NER
   • Save processed results to CSV files

2. Sentiment Analysis Workflow:

   • Uses a branching chain to route reviews to appropriate handling
   • Positive reviews: Generate thank-you message and trip recommendations
   • Negative reviews: Generate apology, improvement plan, and discount code

3. NER Implementation:

   • Uses spaCy's named entity recognition to extract locations
   • Identifies Geopolitical Entities (GPE), Locations (LOC), and Facilities (FAC)
   • Tracks entity types for further analysis

4. Metrics Evaluation:

   • Compares predicted sentiment against ground truth
   • Calculates accuracy, precision, recall, and F1 score
   • Generates visualizations to understand model performance