hiwot7/fintech-review-analytics

I am tackling a real-world scenario: helping Ethiopian banks (CBE, Bank of Abyssinia, and Dashen Bank) improve their fintech apps by turning messy Google Play Store reviews into concrete product strategies. # FinTech Review Analytics Pipeline An end-to-end data engineering and NLP intelligence pipeline built for the 10 Academy Week 2 Challenge. This project automates the extraction, cleaning, sentiment analysis, and structured storage of user feedback for major Ethiopian banking applications (CBE, Bank of Abyssinia, and Dashen Bank) to deliver actionable business insights. ## 🚀 Project Architecture The pipeline consists of 5 distinct phases designed for modularity and scalability: 1. **Extraction:** Automates live review retrieval via `google-play-scraper`. 2. **Refinement:** Cleans, normalizes date formats, and deduplicates records using `pandas`. 3. **Intelligence:** Processes fine-grained sentiment dynamics with a Hugging Face `DistilBERT` transformer model and clusters feedback into key business pillars (Account Access, Transaction Performance, UI/Design, Customer Support). 4. **Storage:** Streams structured data directly into a local relational `PostgreSQL` database utilizing `psycopg2`. 5. **Activation:** Generates visualization metrics to back up financial technology product recommendations. --- ## 📂 Repository Structure ```text fintech_review-analytics/ │ ├── data/ │ └── raw/ │ ├── cleaned_reviews.csv # Phase 2 Output (Refined data) │ └── processed_reviews.csv # Phase 3 Output (NLP features included) │ ├── src/ │ ├── scraper.py # Extraction & Text Cleaning script │ ├── pipeline.py # Hugging Face Intelligence pipeline │ └── db_loader.py # PostgreSQL Schema creation & DB streaming │ ├── requirements.txt # System dependencies └── README.md # Documentation