Product Sentiment Challenge

💬 Product Sentiment Challenge — real Amazon reviews

Teach a network to tell a positive review from a negative one. Each review becomes a bag of words — which words from the list appear in it. Pick the words you think signal sentiment, shape the network, and train (max 50 epochs). Some words carry strong signal, others are neutral decoys. Watch the accuracy, and type your own review below to see it scored live. Score = test MSE × (1 + 0.4% per parameter), up to 3 attempts.

Data: UCI “Sentiment Labelled Sentences” (Kotzias et al., 2015) — 1,000 real Amazon product reviews, 500 positive / 500 negative, seeded 70/30 split.

WORDS (INPUTS)

Which words signal sentiment? Not all do. Click to toggle. 4 selected.

2 hidden layers

Line blue = positive weight, orange = negative; thickness = strength. Watch them shift as it trains.

OUTPUT

Train MSE

—

Test MSE

—

Accuracy

—

🏆 Your Score lower is better

—

test MSE × (1 + 0.4% per parameter)

Test error

Size penalty

• train • test — perfect fit

MSE per epoch — train / test

🔎 Try it on a review

—

negativepositive

Type a review, then train the model to score it.