Datasets:
You need to agree to share your Hugging Face information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
By accessing this dataset, you agree to use it strictly for academic, research, benchmarking, or security evaluation purposes under CC-BY-NC 4.0 terms.
Log in or Sign Up to review the conditions and access this dataset content.
🛡️ OpenShield Multilingual Dataset
Summary: A massive multilingual spam/ham classification dataset containing ~190,000 samples spanning 21 languages formatted in JSON Lines (
.jsonl).
⚡ Access Status: Auto-Approved Gated Access
Instant Access: This repository uses Hugging Face gated access to track research usage. Access requests are set to AUTO-APPROVED. Simply log in, agree to the conditions, and access is granted immediately.
📌 Dataset Overview
OpenShield is a multilingual text classification benchmark designed to evaluate and train spam detection filters, guardrails, and language models across diverse global languages.
- Total Samples: ~190,000
- Format: JSON Lines (
.jsonl) - Languages (21): Arabic (
ar), Bengali (bn), Chinese (zh), English (en), French (fr), German (de), Hindi (hi), Indonesian (id), Japanese (ja), Javanese (jv), Korean (ko), Marathi (mr), Norwegian (no), Portuguese (pt), Punjabi (pa), Russian (ru), Spanish (es), Swedish (sv), Turkish (tr), Ukrainian (uk), Urdu (ur) - Task: Binary Text Classification (
hamvs.spam)
📂 Repository Structure
The files are split per language into dedicated .jsonl files inside the repository:
├── dataset/
│ ├── arabic.jsonl
│ ├── bengali.jsonl
│ ├── chinese.jsonl
│ ├── english.jsonl
│ ├── french.jsonl
│ ├── german.jsonl
│ ├── hindi.jsonl
│ ├── indonesian.jsonl
│ ├── japanese.jsonl
│ ├── javanese.jsonl
│ ├── korean.jsonl
│ ├── marathi.jsonl
│ ├── norwegian.jsonl
│ ├── portuguese.jsonl
│ ├── punjabi.jsonl
│ ├── russian.jsonl
│ ├── spanish.jsonl
│ ├── swedish.jsonl
│ ├── turkish.jsonl
│ ├── ukrainian.jsonl
│ └── urdu.jsonl
🔍 Data Format & Fields
Each entry is a JSON object with the following fields:
| Field | Type | Description |
|---|---|---|
id |
string |
Unique record identifier |
text |
string |
Content of the message or communication |
lang |
string |
ISO 639-1 language code |
label |
int |
Classification label (0: Ham, 1: Spam) |
Sample Record
{"id": "abc-0001", "text": " -- ", "lang": "zh", "label": 0}
🚀 Loading the Dataset
Option 1: Load a Specific Language
from datasets import load_dataset
# Load English subset
dataset_en = load_dataset(
"json",
data_files="https://huggingface.co/datasets/unreadlogs/OpenShield-Dataset/resolve/main/dataset/english.jsonl",
token=True
)
print(dataset_en["train"][0])
Option 2: Load All Languages Combined
from datasets import load_dataset
# Load all language files into a single unified dataset
dataset_all = load_dataset(
"json",
data_files="https://huggingface.co/datasets/unreadlogs/OpenShield-Dataset/resolve/main/dataset/*.jsonl",
token=True
)
print(f"Total samples loaded: {len(dataset_all['train'])}")
⚖️ License & Attribution
This dataset is distributed under the Creative Commons Attribution-NonCommercial 4.0 International (CC-BY-NC 4.0) license.
- Non-Commercial: May not be used for direct commercial exploitation without prior authorization.
- Intended Use: Academic research, security benchmarking, evaluation of anti-spam and content filtering models.
- Takedown & Removal: If you believe any subset includes copyrighted material or personal data you wish removed, please open a thread in the Community Discussions tab.
- Downloads last month
- -