---
title: UQABench
canonical_url: "https://www.modelscope.cn/datasets/OpenStellarTeam/UQABench"
md_url: "https://www.modelscope.cn/datasets/OpenStellarTeam/UQABench.md"
repository: OpenStellarTeam/UQABench
last_updated: 2025-08-19
license: apache-2.0
storage_size: "17 GB"
downloads: 1183
stars: 0
---

# UQABench

> UQABench - OpenStellarTeam 在 ModelScope 开源的数据集。[KDD'25] UQABench: Evaluating User Embedding for Prompting LLMs in Personalized Question Answering [KDD 2025 Accepted (Oral) Paper] Overview The paper link: UQABench: Evaluating User Embedding for Prompting LLMs…

OpenStellarTeam/UQABench 是 ModelScope 魔搭社区上的数据集，存储大小 17 GB，采用 apache-2.0 许可。

- **Repository**: OpenStellarTeam/UQABench
- **License**: apache-2.0
- **Storage size**: 17 GB
- **Downloads**: 1183
- **Stars**: 0
- **Last updated**: 2025-08-19

Source: https://www.modelscope.cn/datasets/OpenStellarTeam/UQABench

---

[KDD'25] UQABench: Evaluating User Embedding for Prompting LLMs in Personalized Question Answering [KDD 2025 Accepted (Oral) Paper]
## Overview
The paper link: [UQABench: Evaluating User Embedding for Prompting LLMs in Personalized Question Answering.](https://arxiv.org/abs/2502.19178) 

Github: https://github.com/OpenStellarTeam/UQABench 

The source data (Kaggle): [Kaggle](https://www.kaggle.com/datasets/liulangmingliu/uqabench) 

## Description
The UQABench is a benchmark dataset for evaluating user embeddings in prompting LLMs for personalized question answering. The standardized evaluation process includes **pre-training**, **fine-tuning**, and **evaluating** stages. We provide the requirements and quick-start scripts for each stage.

The source data are user interactions collected and processed from Taobao. Following previous work, we randomly split the data into 9:1 as the training and test sets. The dataset statistics are summarized as follows:

| Data Split | Total | #Training | #Test |
|---------------|-------------|------------|------------|
| Interaction | 31,317,087 | 28,094,799 | 3,222,288 |

Specifically, the training set serves in the pre-training and fine-tuning (aligning) stages. Then, we design task-specific question prompts based on the test set. We refine the questions, filter out low-quality questions, and eventually get 7,192 personalized Q&A for the evaluating stage.


## Citation
Please cite our paper if you use our dataset.
```
@inproceedings{liu2025uqabench,
  title={UQABench: Evaluating User Embedding for Prompting LLMs in Personalized Question Answering},
  author={Liu, Langming and Liu, Shilei and Yuan, Yujin and Zhang, Yizhen and Yan, Bencheng and Zeng, Zhiyuan and Wang, Zihao and Liu, Jiaqi and Wang, Di and Su, Wenbo and others},
  booktitle={Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2},
  pages={5652--5661},
  year={2025}
}
```