Sms Malicious Benign: 700,000 English Messages for Smishing Detection

~700,000 English SMS messages for binary classification of spam and smishing, created by notd5a and last updated in March 2026. The dataset was iteratively refined through multiple rounds of error analysis to improve data quality for cybersecurity applications.

Use Cases

• Training binary classifiers to distinguish between benign text and malicious smishing attempts
• Benchmarking NLP models on short-form English text messages for spam filtering
• Analyzing linguistic patterns in malicious SMS content to identify common phishing tactics

Strengths

• ~700,000 rows providing a large sample for text classification
• Iterative refinement process based on model error analysis
• Specific focus on smishing (SMS phishing) rather than general email spam

Limitations

• Restricted to English-only messages, limiting global applicability
• Geographic bias toward the US region as indicated by metadata tags
• CC BY-NC 4.0 license prohibits commercial use

Provenance

Source

notd5a via Hugging Face

Collection Method

Iteratively refined through model error analysis

Freshness

Last updated March 2026.

Geography

United States