Update README.md

README.md

@@ -33,6 +33,7 @@ configs:
 
 [![🏆 Leaderboard](https://img.shields.io/badge/🏆_Leaderboard-DataClawBench-red)](https://gtmllab.github.io/DataClaw/)
 [![GitHub](https://img.shields.io/badge/GitHub-GTML--LAB%2FDataClawBench-181717?logo=github)](https://github.com/GTML-LAB-sysu/DataClaw)
+[![arXiv](https://img.shields.io/badge/arXiv-2605.02503-b31b1b)](https://arxiv.org/abs/2605.02503)
 ![Tasks](https://img.shields.io/badge/Tasks-492-blue)
 ![Categories](https://img.shields.io/badge/Categories-7-green)
 
@@ -377,6 +378,13 @@ The current version contains **492** tasks across **7** categories, with an over
 
 > Except for the 39 `international_comparison` tasks, all others are explicitly restricted in the current task spec to use only `./database/`, with no web search.
 
+
+## 📄 Citation
+
+If you use DataClawBench in your research, please cite our paper  
+[**DataClawBench: An Agent Benchmark for Exploratory Real-World Financial Data Analysis**](https://arxiv.org/abs/2605.02503) (arXiv:2605.02503).
+
+
 ## 🙏 Acknowledgements
 
 DataClawBench is jointly released by Prof. Chuan Chen's team at the School of Computer Science, Sun Yat-sen University, and the Southern Weekly Sci-Tech Power Research Center. We sincerely thank the Southern Weekly Sci-Tech Power Research Center for providing invaluable data and tremendous support.