SentiScope

ML-powered sentiment analyzer for product reviews, by Vikranth Reddy

Try the live model

Type a product review below. A real TF-IDF + logistic regression model, trained in Python, exported as weights, running 100% in your browser, scores it.

Try:
—
—

How it works

1. Data
216 labeled product reviews (positive / negative), generated from sentiment-controlled templates.
2. Features
TF-IDF unigrams: each word weighted by how rare it is across reviews.
3. Model
Logistic regression (scikit-learn): learns a weight per word. 5-fold CV accuracy: 100%.
4. Browser
weights.json (vocabulary + idf + coefficients) is shipped to the site; JavaScript re-runs the exact same math: σ(w·x + b).

Tech stack

Python · scikit-learn · NumPy · vanilla JavaScript · static hosting (Vercel)

Files in this repo

train.py: trains the model and exports weights.json · index.html: this demo · vercel.json: security headers