---
license: apache-2.0
---
## 中文说明
# 多标签分类模型（明/清实录） 

本模型基于 [Jihuai/bert-ancient-chinese](https://huggingface.co/Jihuai/bert-ancient-chinese) 预训练权重，可用于对《明实录》和《清实录》中的文本进行多标签分类推理。

## 训练语料

- **来源**：[《朝鲜王朝实录》](https://sillok.history.go.kr/main/main.do)
- **训练样本数**：约30万
- **标签类别数**：194  
- **文本语言**：中文繁体（亦支持中文简体）  

## 评价指标

| 指标             | 基于 Samples | 基于 Label（Micro） |
| --------------- | :----------: | :-----------------: |
| F1              |   0.7095     |       0.6714        |
| Precision       |   0.7692     |       0.7441        |
| Recall          |   0.7000     |       0.6116        |
| Hamming Loss    |            0.0068                  |

> **注**：虽然本模型能大致区分大多数标签，但对于正例极少的标签表现尚不理想。

## 后续工作

相关工作论文coming soon!

## 引用

如需引用，可先引用本页面，仅供学习交流用！
模型链接：<https://huggingface.co/bztxb/shiluBERT>

## 线上使用示例

见HuggingFace Space: [bztxb/InferShilu](https://huggingface.co/spaces/bztxb/InferShilu)
![image/png](https://cdn-uploads.huggingface.co/production/uploads/6687f5437a67e6a352c0195c/QFshR1AKawoLbCxT1p3Qm.png)


## English Version

# Multi-Label Classification Model (The Veritable Records of the Ming/Qing Dynasty)

This model is built on the pretrained weights of [Jihuai/bert-ancient-chinese](https://huggingface.co/Jihuai/bert-ancient-chinese) and is designed for multi-label classification inference on texts from the **Ming Shilu** and **Qing Shilu**.

## Training Corpus

- **Source**: [The Veritable Records of the Joseon Dynasty](https://sillok.history.go.kr/main/main.do)  
- **Training size**: about 0.3 million
- **Number of Labels**: 194  
- **Text Language**: Traditional Chinese (also supports Simplified Chinese)  

## Evaluation Metrics

| Metric        | Sample-based | Label-based (Micro) |
| ------------- | :----------: | :-----------------: |
| F1 Score      |    0.7095    |       0.6714        |
| Precision     |    0.7692    |       0.7441        |
| Recall        |    0.7000    |       0.6116        |
| Hamming Loss  |    0.0068    |         —           |

> **Note**: Although this model can broadly distinguish most labels, its performance on classes with very few positive examples remains suboptimal.

## Future Work

Working paper coming soon!

## Citation

If you wish to cite this work, please refer to this page, welcome any feedback.  
Model link: <https://huggingface.co/bztxb/shiluBERT>


## How to use

See HuggingFace Space: [bztxb/InferShilu](https://huggingface.co/spaces/bztxb/InferShilu)
![image/png](https://cdn-uploads.huggingface.co/production/uploads/6687f5437a67e6a352c0195c/QFshR1AKawoLbCxT1p3Qm.png)