---
license: mit
tags:
- stopwords
- japanese
- nlp
- text-processing
- tokenization
- nagisa
pretty_name: Japanese Stopwords for nagisa
size_categories:
- n<1K
language:
- ja
task_categories:
- text-classification
- feature-extraction
dataset_info:
  features:
  - name: words
    dtype: string
  - name: postags
    dtype: string
  splits:
  - name: nagisa_stopwords
    num_examples: 147
configs:
- config_name: default
  data_files:
  - split: nagisa_stopwords
    path: nagisa_stopwords.csv
---

# Japanese Stopwords for nagisa

This dataset is the Japanese stopwords list built into [nagisa](https://github.com/taishi-i/nagisa) (v0.2.12+). It is published here on Hugging Face for easy access and reproducibility.

## Overview

| | |
|---|---|
| Language | Japanese |
| Size | 147 words |
| Source | [CC-100](https://data.statmt.org/cc-100/), [Wikipedia](https://dumps.wikimedia.org/other/cirrussearch/) |
| License | MIT |

## Dataset Description

This dataset contains **147 frequently used Japanese words** extracted from large-scale corpora. Each word is annotated with its part-of-speech (POS) tag according to nagisa's tokenization rules.

Since nagisa v0.2.12, this stopwords list is included by default and can be accessed directly via `nagisa.stopwords`. The list was expanded from 100 to 147 words, with additional POS categories (代名詞, 接続詞) to improve coverage.

The stopwords are constructed from the following sources:
- [CC-100 dataset](https://data.statmt.org/cc-100/) - Monolingual datasets from Common Crawl
- [Wikipedia](https://dumps.wikimedia.org/other/cirrussearch/) - Japanese Wikipedia dumps

## Usage

### Using with nagisa (Recommended)

Since nagisa v0.2.12, stopwords are built-in. No additional installation is required.

```python
import nagisa

# Tokenize text
text = "日本語のストップワードを簡単に利用できます。"
tokens = nagisa.tagging(text)
print(tokens.words)
# ['日本', '語', 'の', 'ストップ', 'ワード', 'を', '簡単', 'に', '利用', 'でき', 'ます', '。']

# Filter stopwords
words = [word for word in tokens.words if word not in nagisa.stopwords]
print(words)
# ['日本', '語', 'ストップ', 'ワード', '簡単', '利用', '。']
```

### Using with Hugging Face Datasets

```bash
pip install datasets
```

This dataset requires `datasets >= 2.10.0`.

```python
from datasets import load_dataset

dataset = load_dataset("taishi-i/nagisa_stopwords")

# Get stopwords
words = dataset["nagisa_stopwords"]["words"]

# Get the part-of-speech tags
postags = dataset["nagisa_stopwords"]["postags"]
```

### Example: Filtering Stopwords from Text

```python
import nagisa
from datasets import load_dataset

# Load stopwords
dataset = load_dataset("taishi-i/nagisa_stopwords")
stopwords = set(dataset["nagisa_stopwords"]["words"])

# Tokenize and filter
text = "東京で美味しいラーメンを食べました"
tokens = nagisa.tagging(text)
print(tokens.words)
# ['東京', 'で', '美味しい', 'ラーメン', 'を', '食べ', 'まし', 'た']

filtered_words = [word for word in tokens.words if word not in stopwords]
print(filtered_words)
# ['東京', '美味しい', 'ラーメン', '食べ']
```

### Data Fields

| Field | Type | Description |
|-------|------|-------------|
| `words` | string | Japanese word or character |
| `postags` | string | Part-of-speech tag in Japanese |

### Part-of-Speech Tags

The dataset includes the following POS tags:

| POS Tag | English         | Count |
|---------|-----------------|-------|
| 助動詞  | Auxiliary Verb  | 37    |
| 助詞    | Particle        | 35    |
| 動詞    | Verb            | 24    |
| 名詞    | Noun            | 13    |
| 代名詞  | Pronoun         | 13    |
| 接続詞  | Conjunction     | 9     |
| 接尾辞  | Suffix          | 5     |
| 副詞    | Adverb          | 3     |
| 接頭辞  | Prefix          | 2     |
| 連体詞  | Adnominal       | 2     |
| 感動詞  | Interjection    | 2     |
| 形状詞  | Adjectival Noun | 1     |
| 空白    | Whitespace      | 1     |

### Sample Data

| words | postags |
|-------|---------|
| の | 助詞 |
| て | 助詞 |
| に | 助詞 |
| を | 助詞 |
| が | 助詞 |
| は | 助詞 |
| と | 助詞 |
| た | 助動詞 |
| し | 動詞 |
| で | 助詞 |

## Use Cases

- **Text preprocessing**: Remove common words before text analysis
- **Feature extraction**: Reduce noise in bag-of-words or TF-IDF representations
- **Search optimization**: Filter out stopwords from search queries
- **Text classification**: Improve model performance by removing non-informative words


## Related Resources

- [nagisa](https://github.com/taishi-i/nagisa) - Japanese tokenizer based on RNN
- [CC-100](https://data.statmt.org/cc-100/) - Monolingual datasets from Common Crawl
- [Wikipedia Dumps](https://dumps.wikimedia.org/) - Wikipedia database dumps

## Citation

If you use this dataset, please cite:

```bibtex
@misc{nagisa_stopwords,
  author = {Taishi Ikeda},
  title = {Japanese Stopwords for nagisa},
  year = {2023},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/taishi-i/nagisa_stopwords}
}
```