Mobile Review Sentiment Analysis using Machine Learning

End-to-End NLP Project | Logistic Regression | TF-IDF | Business Insights | 87% Accuracy

Project Overview

Customer reviews contain rich insights that go far beyond star ratings. However, manually analyzing thousands of reviews is inefficient and often inaccurate.

This project builds a Machine Learning-based Sentiment Analysis system capable of automatically classifying mobile phone reviews into:

• Positive
• Neutral
• Negative

The model extracts meaningful patterns from textual feedback and transforms them into actionable business insights for product teams, marketers, and decision-makers.

Business Problem

Mobile brands receive massive volumes of customer feedback across regions and platforms. Relying solely on ratings often misrepresents true customer satisfaction.

Key Questions Solved:

1. Can sentiment be accurately predicted from review text alone?

2. Why do some low-rated phones still receive positive feedback?

3. What factors drive negative sentiment?

4. How does sentiment vary across price ranges and regions?

Solution Approach

Data Analysis Pipeline

1. Exploratory Data Analysis

• Sentiment distribution analysis
• Rating vs sentiment comparison
• Regional brand perception
• Price bucket sentiment trends

2. Text Preprocessing

• Lowercasing & regex cleaning
• Stopword removal
• Lemmatization using NLTK

3. Feature Engineering

• TF-IDF Vectorization (5000 features)

4. Model Building

• Logistic Regression classifier
• Stratified Train-Test split

5. Model Evaluation

• Classification report
• Confusion matrix
• Performance analysis

Model Performance

Metric	Result
Accuracy	~87%
Precision	Strong for Positive & Negative
Challenge	Neutral sentiment is hardest to classify

Conclusion: Sentiment can be predicted reliably using textual data alone.

Key Business Insights

• Customer ratings do NOT always reflect true sentiment
• Battery life and performance are the biggest drivers of negative reviews
• Premium phones receive harsher criticism due to higher expectations
• The same brand is perceived differently across regions

These insights can directly support:

• Product improvement strategies
• Pricing decisions
• Market positioning
• Customer experience optimization

Tech Stack

Languages & Libraries

• Python
• Pandas, NumPy
• Scikit-learn
• NLTK
• Matplotlib, Seaborn
• WordCloud

Machine Learning

• TF-IDF Vectorizer
• Logistic Regression

Concepts Applied

• Natural Language Processing (NLP)
• Feature Engineering
• Classification
• Model Evaluation
• Business Analytics

👨‍💻 Author

Soundharya Sundaram Iyer

Github Link: https://github.com/Soundharya09/Mobile-Review-Sentiment-Analysis-using-Machine-Learning

If you like this project, consider giving it a ⭐ on GitHub!