The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

Dataset Card for wisesight1000

Dataset Summary

wisesight1000 contains Thai social media texts randomly drawn from the full wisesight-sentiment, tokenized by human annotators. Out of the labels neg (negative), neu (neutral), pos (positive), q (question), 250 samples each. Some texts are removed because they look like spam. Because these samples are representative of real world content, we believe having these annotaed samples will allow the community to robustly evaluate tokenization algorithms.

Supported Tasks and Leaderboards

word tokenization

Languages

Thai

Dataset Structure

Data Instances

{'char': ['E', 'u', 'c', 'e', 'r', 'i', 'n', ' ', 'p', 'r', 'o', ' ', 'a', 'c', 'n', 'e', ' ', 'ค', '่', 'ะ', ' ', 'ใ', 'ช', '้', 'แ', 'ล', '้', 'ว', 'ส', 'ิ', 'ว', 'ข', 'ึ', '้', 'น', 'เ', 'พ', 'ิ', '่', 'ม', 'ท', 'ุ', 'ก', 'ว', 'ั', 'น', ' ', 'ม', 'า', 'ด', 'ู', 'ก', 'ั', 'น', 'น', 'ะ', 'ค', 'ะ', ' ', 'ว', '่', 'า', 'จ', 'ั', 'ด', 'ก', 'า', 'ร', 'ป', 'ั', 'ญ', 'ห', 'า', 'ส', 'ิ', 'ว', 'ใ', 'น', '7', 'ว', 'ั', 'น', 'ไ', 'ด', '้', 'ร', 'ึ', 'ม', 'ั', '่', 'ย', 'ย', 'ย', 'ย', 'ย', 'ย', 'ย', 'ย', ' ', 'ล', '่', 'า', 'ส', 'ุ', 'ด', 'ไ', 'ป', 'ล', '้', 'า', 'ง', 'ห', 'น', '้', '…', '\n'], 'char_type': [0, 8, 8, 8, 8, 8, 8, 5, 8, 8, 8, 5, 8, 8, 8, 8, 5, 1, 9, 10, 5, 11, 1, 9, 11, 1, 9, 1, 1, 10, 1, 1, 10, 9, 1, 11, 1, 10, 9, 1, 1, 10, 1, 1, 4, 1, 5, 1, 10, 1, 10, 1, 4, 1, 1, 10, 1, 10, 5, 1, 9, 10, 1, 4, 1, 1, 10, 1, 1, 4, 1, 3, 10, 1, 10, 1, 11, 1, 2, 1, 4, 1, 11, 1, 9, 1, 10, 1, 4, 9, 1, 1, 1, 1, 1, 1, 1, 1, 5, 1, 9, 10, 1, 10, 1, 11, 1, 1, 9, 10, 1, 3, 1, 9, 4, 4], 'is_beginning': [1, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0]}
{'char': ['แ', 'พ', 'ง', 'เ', 'ว', '่', 'อ', 'ร', '์', ' ', 'เ', 'บ', 'ี', 'ย', 'ร', '์', 'ช', '้', 'า', 'ง', 'ต', '้', 'น', 'ท', 'ุ', 'น', 'ข', 'ว', 'ด', 'ล', 'ะ', 'ไ', 'ม', '่', 'ถ', 'ึ', 'ง', ' ', '5', '0', ' ', 'ข', 'า', 'ย', ' ', '1', '2', '0', ' ', '😰', '😰', '😰', '์', '\n'], 'char_type': [11, 1, 1, 11, 1, 9, 1, 1, 7, 5, 11, 1, 10, 1, 1, 7, 1, 9, 10, 1, 1, 9, 1, 1, 10, 1, 1, 1, 1, 1, 10, 11, 1, 9, 1, 10, 1, 5, 2, 2, 5, 1, 10, 1, 5, 2, 2, 2, 5, 4, 4, 4, 7, 4], 'is_beginning': [1, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0]}

Data Fields

  • char: characters
  • char_type: character types as adopted from by deepcut
  • is_beginning: 1 if beginning of word else 0

Data Splits

No explicit split is given.

Dataset Creation

Curation Rationale

The dataset was created from wisesight-sentiment to be a word tokenization benchmark that is closer to texts in the wild, since other Thai word tokenization datasets such as BEST are mostly texts from news articles, which do not have some real-world features like misspellings.

Source Data

Initial Data Collection and Normalization

The data are sampled from wisesight-sentiment which has the following data collection and normalization:

  • Style: Informal and conversational. With some news headlines and advertisement.
  • Time period: Around 2016 to early 2019. With small amount from other period.
  • Domains: Mixed. Majority are consumer products and services (restaurants, cosmetics, drinks, car, hotels), with some current affairs.
  • Privacy:
    • Only messages that made available to the public on the internet (websites, blogs, social network sites).
    • For Facebook, this means the public comments (everyone can see) that made on a public page.
    • Private/protected messages and messages in groups, chat, and inbox are not included.
    • Usernames and non-public figure names are removed
    • Phone numbers are masked (e.g. 088-888-8888, 09-9999-9999, 0-2222-2222)
    • If you see any personal data still remain in the set, please tell us - so we can remove them.
  • Alternations and modifications:
    • Keep in mind that this corpus does not statistically represent anything in the language register.
    • Large amount of messages are not in their original form. Personal data are removed or masked.
    • Duplicated, leading, and trailing whitespaces are removed. Other punctuations, symbols, and emojis are kept intact.
    • (Mis)spellings are kept intact.
    • Messages longer than 2,000 characters are removed.
    • Long non-Thai messages are removed. Duplicated message (exact match) are removed.

Who are the source language producers?

Social media users in Thailand

Annotations

Annotation process

[More Information Needed]

Who are the annotators?

The annotation was done by several people, including Nitchakarn Chantarapratin, Pattarawat Chormai, Ponrawee Prasertsom, Jitkapat Sawatphol, Nozomi Yamada, and Attapol Rutherford.

Personal and Sensitive Information

  • The authors tried to exclude any known personally identifiable information from this data set.
  • Usernames and non-public figure names are removed
  • Phone numbers are masked (e.g. 088-888-8888, 09-9999-9999, 0-2222-2222)
  • If you see any personal data still remain in the set, please tell us - so we can remove them.

Considerations for Using the Data

Social Impact of Dataset

  • word tokenization dataset from texts in the wild

Discussion of Biases

  • no guideline is given by the authors on word tokenization

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

Thanks PyThaiNLP community, Kitsuchart Pasupa (Faculty of Information Technology, King Mongkut's Institute of Technology Ladkrabang), and Ekapol Chuangsuwanich (Faculty of Engineering, Chulalongkorn University) for advice. The original Kaggle competition, using the first version of this corpus, can be found at https://www.kaggle.com/c/wisesight-sentiment/

Licensing Information

CC0

Citation Information

Dataset:

@software{bact_2019_3457447,
  author       = {Suriyawongkul, Arthit and
                  Chuangsuwanich, Ekapol and
                  Chormai, Pattarawat and
                  Polpanumas, Charin},
  title        = {PyThaiNLP/wisesight-sentiment: First release},
  month        = sep,
  year         = 2019,
  publisher    = {Zenodo},
  version      = {v1.0},
  doi          = {10.5281/zenodo.3457447},
  url          = {https://doi.org/10.5281/zenodo.3457447}
}

Character type features:

@inproceedings{haruechaiyasak2009tlex,
  title={TLex: Thai lexeme analyser based on the conditional random fields},
  author={Haruechaiyasak, Choochart and Kongyoung, Sarawoot},
  booktitle={Proceedings of 8th International Symposium on Natural Language Processing},
  year={2009}
}

Contributions

Thanks to @cstorm125 for adding this dataset.

Downloads last month
117