ZombitX64/Wisesight-Sentiment-Thai

Dataset Card for Zombitx64 Sentiment Corpus Thai

This dataset card describes the "Zombitx64 Sentiment Corpus Thai," a large-scale, manually curated corpus for Thai sentiment analysis and token classification.

Dataset Details

Dataset Description

The Zombitx64 Sentiment Corpus Thai is a collection of Thai-language social media comments and posts, annotated for sentiment at the sentence or token level. The dataset covers a wide range of topics and emotional expressions, including positive, negative, neutral, mixed, and question classes. It is designed to support research and development in Thai NLP, especially for sentiment analysis and emotion classification.

• Curated by: zombitx64 Co., Ltd.
• Funded by: zombitx64 Co., Ltd.
• Shared by: zombitx64 Co., Ltd.
• Language(s) (NLP): Thai (th)
• License: CC BY-SA 4.0

Dataset Sources

• Repository: github

Uses

Direct Use

• Training and evaluating sentiment analysis models for Thai.
• Research in emotion and opinion mining for Thai social media.
• Benchmarking token-classification models for Thai language.

Out-of-Scope Use

• Not suitable for commercial use without proper attribution.
• Not intended for use in applications requiring highly sensitive or private data filtering.

Dataset Structure

The dataset is provided in JSONL format, where each line is a JSON object with at least the following fields:

• text: The original Thai text.
• sentiment: The sentiment label (e.g., positive, negative, neutral, mixed, question).

Example:

{"text": "มาาาาาา🎉", "sentiment": "positive"}

Loading the Dataset

To load only the JSONL data without including the assets/ folder, use the datasets library directly:

from datasets import load_dataset

ds = load_dataset(
    "json",
    data_files="D:/Github/Wisesight-Sentiment-Thai/sentiment_zombitx64V2.jsonl",
    split="train"
)
print(ds)

Sentiment Distribution

The following chart shows the distribution of sentiment labels in the dataset:

Sentiment	Count
positive	~270,000
neutral	~210,000
question	~88,000
mixed	~45,000
negative	~17,000

Evaluation Metrics

Confusion Matrix

F1 Scores

Per-Class Accuracy

Dataset Creation

Curation Rationale

The dataset was created to address the lack of large-scale, high-quality Thai sentiment datasets for NLP research and applications.

Source Data

Data Collection and Processing

• Data was collected from public social media platforms (e.g., YouTube, Facebook, Twitter).
• Filtering and normalization were performed to remove duplicates, spam, and non-Thai content.
• Manual annotation was conducted by native Thai speakers following detailed guidelines.

Who are the source data producers?

• The original data was produced by Thai social media users.
• Annotation was performed by trained linguists and annotators at zombitx64.

Annotations

Annotation process

• Annotation guidelines were developed to ensure consistency.
• Each comment was labeled with one of the sentiment classes.
• Quality control included double annotation and adjudication for ambiguous cases.

Who are the annotators?

• Native Thai speakers with experience in linguistics and sentiment analysis.

Personal and Sensitive Information

• The dataset was anonymized to remove personal identifiers.
• Only public comments were included; private or sensitive data was excluded to the best of the curators' ability.

Bias, Risks, and Limitations

• The dataset may reflect biases present in Thai social media.
• Some topics or dialects may be underrepresented.
• Sentiment annotation is inherently subjective; disagreements may exist.

Recommendations

Users should be aware of potential biases and limitations. It is recommended to supplement this dataset with additional data for production use in sensitive applications.

Citation

BibTeX:

@dataset{zombitx64_sentiment_corpus_thai,
  title = {Zombitx64 Sentiment Corpus Thai},
  author = {zombitx64 Co., Ltd.},
  year = {2025},
  license = {CC BY-SA 4.0}
}

APA: zombitx64 Co., Ltd. (2025). Zombitx64 Sentiment Corpus Thai [Data set].

Glossary

• Token-classification: Assigning a label to each token or segment in the text.
• Sentiment: The emotional tone (positive, negative, neutral, mixed, question) expressed in the text.

More Information

For questions or collaboration, please contact zombitx64 Co., Ltd.

Dataset Card Authors

• zombitx64 Data Team