README.md

---
language:
- zh
- en
license: apache-2.0
size_categories:
- 100M<n<1B
pretty_name: AL-GR Item Titles
tags:
- item-metadata
- text
- ner
- anonymized
- e-commerce
- recommendation
- generative-retrieval
task_categories:
- text-retrieval
dataset_info:
- config_name: default
  splits:
  - name: train
    num_examples: 255000000
---

# AL-GR/Item-Info: Anonymized Item Titles

<!-- Paper: [FORGE: Forming Semantic Identifiers for Generative Retrieval in Industrial Datasets](https://huggingface.co/papers/2509.20904)
Code: https://github.com/selous123/al_sid
Project Page: https://huggingface.co/AL-GR -->

## Dataset Summary

This repository, `AL-GR/Item-Info`, is a companion dataset to the `AL-GR` generative recommendation ecosystem. It provides a crucial mapping from the abstract item identifiers (`base62_string`) to their corresponding **anonymized text titles**.

The dataset contains approximately 255 million item-title pairs. A key feature of this dataset is that the `title` column has been carefully processed for **user privacy and information protection**.

This dataset can be used for:
- Understanding the semantic meaning behind the abstract item IDs.
- Creating text-based features for recommendation models.
- Debugging and analyzing recommendation results by looking up the item's category.

## Anonymization and Data Curation

The `title` column does not contain the original, full-length product titles. To protect user data and remove sensitive information (such as brand names, specific sellers, or promotional text), a multi-step anonymization process was applied:

1.  **Named Entity Recognition (NER)**: An NER model was first run on the original titles to identify key entities like product category, attributes, materials, etc.
2.  **Entity Selection & Truncation**: Only the most representative and generic entities were preserved. The result was then truncated to a short, descriptive string (e.g., "运动裤 长裤" for "Men's Nike Tech Fleece Jogger Sweatpants in Dark Grey Heather").

This process ensures that the title retains its core semantic meaning while being fully anonymized.

## How to Use

The dataset is stored as a single large CSV file and can be easily loaded using the `datasets` library.

```python
from datasets import load_dataset

# Load the dataset from the Hugging Face Hub
# NOTE: Replace [your-username] with the actual username
dataset = load_dataset("AL-GR/Item-Info")

# Inspect a sample
sample = dataset['train'][0]
print(sample)

# Expected output:
# {
#   'base62_string': 'Tp2BF',
#   'title': '运动裤 长裤'
# }
```

## Dataset Structure

### Data Fields

- `base62_string` (string): The unique, 5-character identifier for the item. This ID is consistent across the entire `AL-GR` ecosystem.
- `title` (string): The **anonymized and processed** item title. It typically contains key category or attribute words.

### Data Splits

| Split      | Number of Samples      |
|------------|------------------------|
| `train`    | ~255,000,000           |


## Citation

If you use this dataset in your research, please cite:

<!-- ```bibtex
@misc{fu2025forgeformingsemanticidentifiers,
      title={FORGE: Forming Semantic Identifiers for Generative Retrieval in Industrial Datasets}, 
      author={Kairui Fu and Tao Zhang and Shuwen Xiao and Ziyang Wang and Xinming Zhang and Chenchi Zhang and Yuliang Yan and Junjun Zheng and Yu Li and Zhihong Chen and Jian Wu and Xiangheng Kong and Shengyu Zhang and Kun Kuang and Yuning Jiang and Bo Zheng},
      year={2025},
      eprint={2509.20904},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2509.20904}, 
}
``` -->

## License

This dataset is licensed under the Apache-2.0 License.