---
license: cc-by-4.0
tags:
  - passphrase
  - security
  - password
language:
  - en
task_categories:
  - text-generation
---

# MASCARA Data

Pre-trained probability tables used in the MASCARA passphrase guessability experiment.

## Contents

`mascara-data.zip` extracts to:

| File | Size | Description |
|---|---|---|
| `biprob.pkl` | 236M | Bigram probability table trained on Wikipedia-5 corpus. Format: `"word1 word2" → float`. 7,584,975 pairs. |
| `bigram.pickle` | 89M | Raw bigram count table. Format: `word → defaultdict(word → count)`. 496,161 keys. |
| `unigram-5p.pkl` | 6.5M | Unigram probability table (requires NLTK). |

## Usage

```bash
# Download
hf download wei192026/mascara-data mascara-data.zip \
    --repo-type dataset \
    --local-dir data/

# Extract
cd data/ && unzip mascara-data.zip
# Results in: biprob.pkl, bigram.pickle, unigram-5p.pkl
These files are used by src/attack_models.py as the Wiki-5 bigram attacker (Model 6 in the min-auto framework).
```

## Source
Data originates from the MASCARA paper:


@inproceedings{mukherjee-2023-memorable-passphrase,
  author    = {Avirup Mukherjee and Kousshik Murali and Shivam Kumar Jha and Niloy Ganguly and Rahul Chatterjee and Mainack Mondal},
  title     = {{MASCARA: Systematically Generating Memorable And Secure Passphrases}},
  booktitle = {ACM ASIACCS 2023},
  year      = {2023}
}


---