SMS Spam Robustness

GitHub repository

Message input

0 characters

Examples:

Each condition changes at most one eligible word per message.

Seed: 42. URLs, email addresses and phone numbers are preserved.

Text comparison

No results yet.

Original

—

Modified

—

Normalized

—

Normalization log

Uses only the training vocabulary. Ambiguous words, likely names and abbreviations are left unchanged.

  • No results yet.

Model comparison

Apply a change to compare the trained models.

ModelOriginal predictionModified prediction
Majority baseline——
A: Word TF-IDF + Logistic Regression——
B: Character TF-IDF + Logistic Regression——
C: Normalization + Word TF-IDF + Logistic Regression——