README.md

---
license: apache-2.0
language:
- zh
tags:
- finance
---
# Dataset Card for CRAFT

[![arXiv](https://img.shields.io/badge/arXiv-2508.01302-b31b1b.svg?logo=arxiv)](https://arxiv.org/abs/2508.01302)
[![GitHub](https://img.shields.io/badge/GitHub-CRAFT_&_KEDAS-blue?logo=github)](https://github.com/JamyDon/LTE)
[![License](https://img.shields.io/badge/License-Apache%202.0-blue.svg?logo=apache)](LICENSE)

<!-- Provide a quick summary of the dataset. -->

This is CRAFT, a dataset for **C**hinese **R**eal-time statistics **A**nd **F**inance knowledge edi**T**ing). CRAFT supports real-time data curation with a [fully automated pipeline](https://github.com/JamyDon/CRAFT-KEDAS/tree/main/CRAFT).

This repository contains the CRAFT dataset curated in 25Q1.

## Dataset Details

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

- **Curated by:** The CRAFT&KEDAS team.
- **Language(s) (NLP):** Chinese
- **License:** Apache-2.0

### Dataset Sources

<!-- Provide the basic links for the dataset. -->

- Monthly statistical reports from the [National Bureau of Statistics of China](https://data.stats.gov.cn/) via the [`cn-stats` API](https://github.com/songjian/cnstats).
- Annual financial statements of publicly listed Chinese companies via the [`AKShare` API](https://github.com/akfamily/akshare).
- Commonsense data from [C3](https://dataset.org/c3/).

## Uses

<!-- Address questions around how the dataset is intended to be used. -->

Real-time knowledge editing. Evaluates Edit Success, Locality, and Portability.

## Dataset Structure

<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->

```json
{
  "case_id": "an integer ID",
  "subject": [
    "related subject 1",
    "related subject 2"
  ],
  "prompt": [
    "prompt 1",
    "prompt 2"
  ],
  "target_new": [
    "new target 1",
    "new target 2"
  ],
  "portability": {
    "Subject_Aliasing": [
      {
        "prompt": "subject aliasing query 1",
        "ground_truth": [
          "subject aliasing answer 1"
        ]
      },
      {
        "prompt": "subject aliasing query 2",
        "ground_truth": [
          "subject aliasing answer 2"
        ]
      }
    ],
    "Reasoning": [
      {
        "prompt": "reasoning query",
        "ground_truth": [
          "reasoning answer"
        ]
      }
    ]
  },
  "locality": {
    "Relation_Specificity": [
      {
        "prompt": "relation specificity query 1",
        "ground_truth": [
          "relation specificity answer 1"
        ]
      },
      {
        "prompt": "relation specificity query 2",
        "ground_truth": [
          "relation specificity answer 2"
        ]
      }
    ],
    "common_sense": [
      {
        "prompt": "common sense query 1",
        "ground_truth": [
          "common sense answer 1"
        ]
      },
      {
        "prompt": "common sense query 2",
        "ground_truth": [
          "common sense answer 2"
        ]
      }
    ]
  }
}
```

## Citation

If you find our work useful, feel free to cite our paper:

```bib
@misc{tang2025aligninglanguagemodelsrealtime,
      title={Aligning Language Models with Real-time Knowledge Editing}, 
      author={Chenming Tang and Yutong Yang and Kexue Wang and Yunfang Wu},
      year={2025},
      eprint={2508.01302},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2508.01302}, 
}
```