SolarisCipher/hk_content_corpus

HK Content Corpus (Cantonese & Traditional Chinese)

This dataset contains eight cleaned source-specific corpora of Hong Kong Cantonese and Traditional Chinese text, crawled from public websites and platforms.

It was initially created for the experiments reported in https://doi.org/10.1145/3744341 which study the effect of diglossia on Hong Kong language modeling.

Each file stores plain UTF-8 text, where each record occupies one line, and blank lines serve as separators.

This dataset is also available at Zenodo: https://doi.org/10.5281/zenodo.16882351

We only change file extension from .corpus to .csv and add header row here for HuggingFace's dataset viewer function

👉 This cleaned corpus is derived from a larger MySQL database used to store raw text during the data collection stage.
If you need the original database for reprocessing or reproduction, please refer to:
https://huggingface.co/datasets/SolarisCipher/hk_content_corpus_mysql

Files

Filename	Description	SHA256 hash (without header row, same content as corpus file@Zenodo)
appledaily_article_dedup.csv	Apple Daily news articles	0b6ad22b7a73230fd0e44af904c0cff1773cc871417f6ad3af11a783564bca15
hkcitizenmedia_article_dedup.csv	HK Citizen Media articles	373a5b369d5e402e58760861e2bff2e618e7ee7fd4494988a39be1156f9dba84
hkcnews_article_dedup.csv	Hong Kong Citizen News	67f909cfaf7d7a67df1dc79448f24622ec525a31b73b2e3293fbaad147470e69
inmedia_article_dedup.csv	InMedia.hk articles	135251f3e8ba7587018b7b2814b9ea5bbebf69c98adca73d3ea4c9f9b5571957
lihkg_posts_dedup_demoji_128.csv	Lihkg forum posts (emoji removed)	8e5e6de9c219aeccdaf13e9162b00cce2eeb7f595023a7bf19d4b5660395a3ee
openrice_review_dedup_demoji.csv	OpenRice user reviews	dd5835a7effe49bb96a31e0c0eab43dea17ef23b3e1d9cefdc186aca276897ce
thestandnews_article_dedup.csv	Stand News articles	847ef0f5809481caf4bf21100e4b34207513d724dafb9f42b162ef49079e7dba
wiki_hk.csv	Wikipedia (zh-hk)	bd33008802797b33df8484cf1113be6a0b38547fe13515f5c4edbf4ccad270db

Intended Uses

• Language model pretraining and finetuning
• Hong Kong Cantonese/Traditional Chinese linguistic modeling
• Downstream tasks such as classification or generation

NOTE: HKNSL became effective since 2020-6-30, which can create bias on user content created afterwards. Those portion of data should be used with caution.

Citation

If you use this dataset, please cite the following paper:

@article{Yung2025HKDiglossia,
  author    = {Yung, Yiu Cheong and Lin, Ying-Jia and Kao, Hung-Yu},
  title     = {Exploring the Effectiveness of Pre-training Language Models with Incorporation of Diglossia for Hong Kong Content},
  journal   = {ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP)},
  volume    = {24},
  number    = {7},
  pages     = {71:1--71:16},
  year      = {2025},
  publisher = {Association for Computing Machinery},
  doi       = {10.1145/3744341}
}

and optionally also cite the dataset DOI:

@dataset{yung_2025_16882351,
  author       = {Yung, Yiu Cheong},
  title        = {HK Content Corpus (Cantonese \& Traditional Chinese)},
  month        = aug,
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.16882351},
  url          = {https://doi.org/10.5281/zenodo.16882351},
}