Model Results
Validation metrics and sentiment classification performance.
Back to Projects
ShippedNLP & Analytics
App Review Sentiment Analyzer
BERT-powered sentiment for Google Play reviews
Overview
A sentiment analysis tool fine-tuned on BERT (96% validation accuracy) that classifies app reviews as positive, neutral, or negative — with a Gradio interface for single-review and bulk Play Store analysis.
The Problem
Product teams need fast, reliable sentiment signals from large volumes of app store reviews — beyond star ratings alone.
The Solution
Fine-tuned BERT with confidence scores, bulk scraping from Play Store URLs, distribution charts, and CSV export for downstream analytics.
Architecture
1
Model Layer
Fine-tuned bert-base-uncased for three-class sentiment with confidence scoring.
2
Inference API
Tokenization and batch prediction for single and bulk review inputs.
3
Data Acquisition
Play Store scraper extracts reviews from package names or full store URLs.
4
Gradio UI
Web interface with charts, examples, and CSV export for analysis workflows.
Technology
BERTPyTorchHugging Face TransformersGradioGoogle Play ScraperXGBoost
Features
- Single-review sentiment with confidence scores
- Bulk analysis up to 200 reviews per app
- Play Store URL and package name parsing
- Sentiment distribution visualizations
- Star-rating breakdown charts
- CSV export for further analysis
Timeline
1
Dataset & Baselines
BiLSTM, XGBoost, and BERT model comparison
2
BERT Fine-tuning
96% validation accuracy on three-class labels
3
Gradio Interface
Interactive UI for single and bulk review analysis
4
Play Store Integration
Scraper and visualization pipeline for live apps
Platform Screens
Ready to build institutional intelligence?
Let's discuss how Dreamskrin can architect your next platform.
Book Strategy Call