---
dataset_info:
  features:
  - name: anchor
    dtype: string
  - name: positive
    dtype: string
  splits:
  - name: train
    num_bytes: 1499094
    num_examples: 43447
  - name: test
    num_bytes: 377209
    num_examples: 10881
  download_size: 584371
  dataset_size: 1876303
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---

# Typosquat Embedding Dataset

## Dataset Summary

This dataset is designed for training embedding models to recognize typosquatting within a domain corpus. It consists of pairs of legitimate and typosquatted domains for use in similarity learning, enabling models to identify subtle domain alterations.

The dataset is formatted for embedding-based training, specifically useful for contrastive learning techniques or other tasks where domain similarity is a key factor. 

## Supported Tasks and Leaderboards

**Embedding Training**: The primary task supported by this dataset is contrastive learning to create embeddings for typosquatting detection. The dataset can be used to train a similarity model, such as a dual-encoder, where each instance is a pair of legitimate and potentially typosquatted domains.

## Languages

This dataset includes a multilingual set of domains, reflecting the diversity of internet domains globally.

## Dataset Structure

### Data Instances

Each instance in the dataset consists of two domains:

- **anchor**: The legitimate domain.
- **positive**: A version of the domain with minor alterations that may represent typosquatting.

An example from the dataset is as follows:

```json
{
  "anchor": "e-volution.ai",
  "positive": "e-volutiọn.ai"
}
```
The anchor and positive columns are both strings representing domains. 
The "positive" domain is a variation created by intentional typosquatting techniques (e.g., homoglyphs or character substitution).

### Data Splits

The dataset is structured to be used for embedding model training and evaluation:

  - Split: Train
    Number of Instances: 43,447
  - Split: Test
    Number of Instances: 10,881

## Dataset Creation

### Data Generation
The domain pairs were generated using [ail-typo-squatting](https://github.com/typosquatter/ail-typo-squatting) 
Data processing includes balancing positive and negative samples to ensure even representation.

### Dataset Usage

This dataset is suitable for cybersecurity applications focusing on typosquatting detection. It can be used to train and evaluate embedding-based models designed to identify domains that may have been manipulated for malicious purposes, supporting efforts in online safety and domain monitoring.