p988744/eland-sentiment-zh-data

---
license: apache-2.0
language:
- zh
task_categories:
- text-classification
tags:
- sentiment-analysis
- finance
- chinese
- taiwan
- stock-market
size_categories:
- 1K<n<10K
---

# Eland Sentiment Chinese Financial Dataset

A Chinese financial sentiment analysis dataset for Taiwan stock market text.

## Dataset Description

This dataset contains 1,299 annotated samples of Chinese financial text from Taiwan stock market forums and news. It supports three sentiment analysis tasks:

1. **Overall Sentiment** - Classify the overall sentiment of the text
2. **Entity Sentiment** - Classify sentiment towards specific entities (companies, products)
3. **Opinion Sentiment** - Classify sentiment of specific opinions in text

## Dataset Statistics

| Split | Samples |
|-------|---------|
| Train | 999 |
| Test | 300 |
| **Total** | **1,299** |

### Task Distribution

| Task | Train | Test |
|------|-------|------|
| Overall Sentiment | 333 | 100 |
| Entity Sentiment | 333 | 100 |
| Opinion Sentiment | 333 | 100 |

### Sentiment Distribution (Train)

| Sentiment | Count | Percentage |
|-----------|-------|------------|
| Positive (正面) | 434 | 43.4% |
| Neutral (中立) | 350 | 35.0% |
| Negative (負面) | 215 | 21.5% |

## Data Format

Each sample is a JSON object with the following fields:

### Overall Sentiment Task
```json
{
"text": "台積電今日股價大漲...",
"overall": "正面",
"task": "overall",
"source": "forum_01"
}
```

### Entity Sentiment Task
```json
{
"text": "台積電今日股價大漲...",
"overall": "正面",
"entity": "台積電",
"entity_sentiment": "正面",
"task": "entity",
"source": "forum_01"
}
```

### Opinion Sentiment Task
```json
{
"text": "台積電今日股價大漲...",
"overall": "正面",
"opinion": "股價上漲代表市場看好",
"opinion_sentiment": "正面",
"agrees_with_text": true,
"task": "opinion",
"source": "forum_01"
}
```

## Labels

- **正面** (Positive)
- **中立** (Neutral)
- **負面** (Negative)

## Usage

### Load with Datasets Library

```python
from datasets import load_dataset

dataset = load_dataset("p988744/eland-sentiment-zh-data")

# Access splits
train_data = dataset["train"]
test_data = dataset["test"]

# Example
print(train_data[0])
```

### Load Manually

```python
import json

with open("train.jsonl", "r", encoding="utf-8") as f:
train_data = [json.loads(line) for line in f]
```

## Associated Model

This dataset was used to train [p988744/eland-sentiment-zh](https://huggingface.co/p988744/eland-sentiment-zh), which achieves **89.38% Reliability** on the RGL benchmark.

## License

Apache 2.0

## Citation

```bibtex
@misc{eland-sentiment-zh-data,
author = {Eland AI},
title = {Eland Sentiment: Chinese Financial Sentiment Dataset},
year = {2025},
publisher = {HuggingFace},
url = {https://huggingface.co/datasets/p988744/eland-sentiment-zh-data}
}
```