Customer Review Analysis with NLP: Sentiment Analysis and Topic Modeling using Python

Customer reviews contain valuable information about user satisfaction, service quality, and business performance. However, manually analyzing thousands of reviews is inefficient and often impossible at scale.

In this project, I developed a Natural Language Processing (NLP) pipeline to automatically analyze customer reviews and extract actionable business insights from unstructured text data.

The full project repository is available here:
GitHub: https://github.com/ivanmartiolme/Proyecto-Master-DataScience-Evolve-IvanMartinez

Project Objectives

The main goals of this project were:

• Clean and preprocess customer review text
• Perform sentiment analysis to classify reviews as positive or negative
• Identify recurring discussion topics automatically
• Analyze sentiment by topic
• Generate business insights from textual data

Tools and Technologies

The project was developed using:

• Python
• Pandas
• NumPy
• Scikit-learn
• Transformers (Hugging Face)
• Datasets (Hugging Face)
• Matplotlib
• Jupyter Notebook

Methodology

1. Data Collection and Preparation

Customer reviews were loaded into a structured DataFrame for analysis. A full text preprocessing pipeline was applied, including:

• Lowercasing
• URL removal
• Punctuation and number removal
• Whitespace normalization

This ensured better model performance and cleaner topic extraction.

2. Sentiment Analysis

A pretrained Transformer-based sentiment analysis model was used to classify each review as positive or negative.

This allowed measurement of overall customer satisfaction and helped identify problematic areas.

3. Topic Modeling

To uncover the most frequent themes discussed by customers, TF-IDF vectorization and Non-negative Matrix Factorization (NMF) were applied.

This revealed recurring topics such as:

• Airport parking and transportation
• Customer service experience
• Pricing and additional fees
• Refund and cancellation issues

4. Insight Generation

Combining sentiment analysis with topic modeling made it possible to detect which topics generated the most negative feedback and where service improvements could be prioritized.

Key Learnings

This project demonstrated how NLP techniques can transform large volumes of unstructured customer feedback into meaningful business intelligence.

It also reinforced practical skills in:

• Text preprocessing
• Transformer-based sentiment analysis
• Topic modeling
• Data storytelling

Conclusion

Customer review analysis using NLP can significantly improve decision-making and customer experience strategies.

This project was developed as part of the Master in Data Science at Evolve.