Project NLP | Business Case: Automated Customer Reviews

The goal of this project is to develop a product review system. That can classify review sentiment to Positive/Neutral/Negative, cluster the reviews by product categories and write product summaries into recommendation articles.

For this project a dataset with Amazon reviews is used, more specifically reviews from the Video Games category. This data set can be found here 👉 https://amazon-reviews-2023.github.io/

Full project guidelines can be found here 👇
https://github.com/ironhack-labs/project-nlp-business-case-automated-customers-reviews-v2

📊 Best Results

Best-performing model for review sentiment classification was RoBERTa-base:

• Accuracy: 0.875
• F1 Score (Macro Avg): 0.734
• Precision (Macro Avg): 0.719
• Recall (Macro Avg): 0.766
• Weighted F1 Score: 0.885

🔍 Clusters found

• Keyboards & Mice
• Headsets & Audio
• Games
• Controllers

🤗 Models and deployment on HuggingFace

👉 https://huggingface.co/spaces/DaanBooy/games_and_accessories_reviews 👈

What you can do:

• Classification: Upload a review and see its sentiment classification (RoBERTa-base).
• Clustering: Explore product category clusters.
• Summaries: Read AI-generated articles highlighting top products, main complaints, and the worst product per category.

The model files generated in this project can be found here: 👇

• https://huggingface.co/DaanBooy/review-sentiment-distilbert-base-uncased
• https://huggingface.co/DaanBooy/review-sentiment-bert-base-uncased
• https://huggingface.co/DaanBooy/review-sentiment-roberta-base

⚙️ How to Run & Reproduce Results

Clone the repository

git clone <your-repo-url>
cd <repo-folder>

Install dependencies

pip install -r requirements.txt

Data Preparation
Run data_prep.ipynb to preprocess the Amazon Video Games review dataset.

Train Models

• review_classification_dbu.ipynb → DistilBERT
• review_classification_bbu.ipynb → BERT-base
• review_classification_rb.ipynb → RoBERTa-base

Evaluate Models
Use model_comparison_viz.ipynb to visualize results (precision, recall, F1, confusion matrix).

Clustering

• Run review_clustering_sample.ipynb for exploratory clustering.
• Run review_clustering_full.ipynb to cluster the full dataset into 4 categories.

Summarization
Execute review_summarization.ipynb to generate product recommendation articles.

Deployment
Use model_deployment_HF.ipynb to deploy the system to Hugging Face Spaces.

Notes:
To ensure you have enough RAM to run notebooks, use Google Colab on GPU T4 (High-RAM).
To run review_summarization.ipynb, make sure you create a .env file in your project folder/Google Drive containing your OpenAI API key:

OPENAI_API_KEY=your_api_key_here

---

📂 Repository Contents

File Name	Description
requirements.txt	Lists all Python dependencies needed to run and reproduce the entire project
data_prep.ipynb	Used to prepare review data for the rest of the project
review_classification_dbu.ipynb	Notebook with distilbert-base-uncased model training
review_classification_bbu.ipynb	Notebook with bert-base-uncased model training
review_classification_rb.ipynb	Notebook with roberta-base model training
model_comparison_viz.ipynb	Notebook used to visualize model results
review_clustering_sample.ipynb	Notebook used to cluster sample of data to inspect dataset
review_clustering_full.ipynb	Notebook used to full cluster the entire dataset
review_summarization.ipynb	Notebook used to make the generative review summarization articles
model_deployment_HF	File used to deploy to HuggingFace space
Project NLP _ Business Case_ Automated Customer Reviews - Final report.pdf	PDF containing the final report on this project
Presentation.pptx	Powerpoint Presentation of this project

---