Text Generation: Pre-training or fine-tuning language models on Traditional Chinese news text
Text Classification: News categorization, sentiment analysis
Summarization: News headline generation, article summarization
Information Extraction: Named entity recognition, event extraction

Dataset Structure

Data Fields

Field	Type	Description
`id`	string	Unique article identifier
`title`	string	Article headline (Traditional Chinese)
`content`	string	Full article text
`publish_date`	string	Publication date (YYYY-MM-DD)
`source`	string	Source identifier (tor/van)
`category`	string	News category
`url`	string	Original article URL
`word_count`	int	Character count
`language`	string	Language code

Data Splits

Split	Number of Examples
train	1,070,292

Example

{
    "id": "tor_20240115_hk-gaa1",
    "title": "政府宣布新措施",
    "content": "【明報專訊】政府今日宣布...",
    "publish_date": "2024-01-15",
    "source": "mingpao_canada_tor",
    "category": "港聞",
    "url": "https://www.mingpaocanada.com/tor/htm/News/20240115/HK-gaa1_r.htm",
    "word_count": 1234,
    "language": "yue"
}

Usage

Loading the Dataset

from datasets import load_dataset

# Load the full dataset
ds = load_dataset("hon9kon9ize/mingpao-canada-news")

# Access data
print(ds["train"][0])

# Filter by date
recent = ds["train"].filter(lambda x: x["publish_date"] >= "2024-01-01")

# Filter by edition (Toronto vs Vancouver)
toronto = ds["train"].filter(lambda x: "tor" in x["source"])
vancouver = ds["train"].filter(lambda x: "van" in x["source"])

Using with Pandas

import pandas as pd
from datasets import load_dataset

ds = load_dataset("hon9kon9ize/mingpao-canada-news")
df = ds["train"].to_pandas()

# Group by year
df["year"] = pd.to_datetime(df["publish_date"]).dt.year
print(df.groupby("year").size())

# Group by category
print(df.groupby("category").size())

Dataset Creation

Source Data

Articles were collected from Ming Pao Canada's online archive, covering all available news sections from both Toronto and Vancouver editions.

Processing

HTML content extraction using BeautifulSoup
Unicode normalization (NFC) for Traditional Chinese
Boilerplate removal (navigation, headers, footers)
Content deduplication via MD5 hashing
Quality filtering (minimum content length)

Curation Rationale

This dataset was created to provide a comprehensive Traditional Chinese news corpus for:

Training language models for Cantonese/Traditional Chinese
Research on Chinese Canadian media coverage
NLP tasks requiring Traditional Chinese text
Historical news analysis

Considerations for Using the Data

Social Impact

This dataset contains news articles covering Canadian and international current events from a Chinese Canadian perspective. Users should be aware that:

News content may reflect editorial perspectives
Historical coverage may contain outdated information
Some topics may be politically sensitive

Biases

Single source (Ming Pao Canada) may have editorial biases
Coverage focuses on news relevant to Chinese Canadian readers
Time period may over/under-represent certain events

Limitations

Some articles may have extraction errors
Original images and multimedia not included
Very short articles (< 50 characters) excluded

Copyright & License

⚠️ Copyright Notice

All article content in this dataset is copyrighted by Ming Pao Canada (明報加拿大).

This dataset is provided for research and educational purposes only. By using this dataset, you agree to:

Not redistribute the content for commercial purposes
Credit Ming Pao Canada as the original copyright holder
Use only for non-commercial research (e.g., NLP research, language model training for academic purposes)
Comply with applicable copyright laws in your jurisdiction

The dataset compiler does not claim any ownership over the original content and provides this dataset under fair use principles for research purposes.

If you are the copyright holder and have concerns about this dataset, please contact us.

Citation

@dataset{mingpao_canada_news_2026,
  title = {Ming Pao Canada Full Archive Corpus},
  author = {hon9kon9ize},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/hon9kon9ize/mingpao-canada-news},
  note = {Content copyrighted by Ming Pao Canada}
}

Acknowledgements

Ming Pao Canada (明報加拿大) - Original content copyright holder

Downloads last month: 9