Game setup
Choose a mode, dataset, difficulty, timing, and accessibility preferences.
Performance dashboard
Review accuracy, class metrics, confidence, timing, and error patterns.
Round review
Inspect each answer, explanation, confidence level, response time, and model disagreement.
| # | Message | Your label | Correct | Confidence | Time | Points |
|---|---|---|---|---|---|---|
| No completed rounds yet. | ||||||
Custom text laboratory
Enter a sentence, inspect a simple prediction, and add it to your private practice set.
Achievements and leaderboard
Progress is saved locally in your browser.
Local leaderboard
| Date | Mode | Score | Accuracy |
|---|---|---|---|
| No saved sessions. | |||
How to play
1. Read carefully
Look for emotional words, negation, contrasts, factual language, and sarcastic clues.
2. Classify and rate confidence
Choose positive, neutral, or negative. Add confidence and optional intensity.
3. Learn from feedback
Review explanations, model probabilities, class metrics, and saved difficult examples.
Sentiment classification guide
Positive sentiment
Positive text expresses satisfaction, praise, gratitude, enthusiasm, comfort, or recommendation. Look beyond isolated words and identify the writer's overall judgment.
Neutral sentiment
Neutral text usually reports facts, schedules, specifications, questions, or balanced observations. Emotional vocabulary may appear without creating a clear overall opinion.
Negative sentiment
Negative text expresses criticism, disappointment, frustration, failure, delay, or risk. The final conclusion often matters more than an earlier compliment.
Negation
Words such as not, never, no longer, and cannot can reverse polarity. “Not bad” is commonly positive, while “not helpful” is negative.
Mixed sentiment
Some messages contain both praise and criticism. Classify the dominant conclusion, recommendation, or strongest consequence rather than simply counting positive words.
Sarcasm
Sarcasm uses literal praise to communicate criticism. Contradictions, quotation marks, exaggeration, and implausible gratitude can reveal the intended meaning.
Frequently asked questions
What is sentiment classification?
It assigns an emotional label, commonly positive, neutral, or negative, to text.
Why can neutral messages be difficult?
Factual messages may contain emotional words while expressing no clear judgment.
How does confidence affect scoring?
Correct high-confidence answers earn bonuses. Incorrect high-confidence answers receive larger penalties.
What is the simulated AI model?
It is a lightweight browser demonstration using sentiment clues and controlled randomness.
What does the confusion matrix show?
It compares actual labels with player predictions for every sentiment class.
What is macro F1?
It averages class-level F1 scores so every sentiment contributes equally.
Can I add my own examples?
Yes. Custom examples are stored locally and can be exported as CSV.
Does the game send text to a server?
No. Gameplay, reports, and custom examples run locally in your browser.
How is the daily challenge selected?
A date-based seed creates the same local challenge for that calendar day.