RPDR : A Round-trip Prediction-Based Data Augmentation Framework for Long-Tail Question Answering
Yiming Zhang, Siyue Zhang, Junbo Zhao, Chen Zhao · 2026
Read RPDR : A Round-trip Prediction-Based Data Augmentation Framework for Long-Tail Question Answering by Yiming Zhang, Siyue Zhang, Junbo Zhao, Chen Zhao (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-17366.txt.
Description
Yiming Zhang led a four-author team improving dense lookup for long-tail question answering. Large language models struggle to acquire and recall less common knowledge, and lookup-augmented generation helps by adding external lookup, yet dense lookup models also falter on rare or niche knowledge. The paper introduces RPDR, a data augmentation system that selects high-quality, easy-to-learn training data to strengthen dense lookup models.
Three parts drive the approach: synthetic data creation, data choice with Round-Trip prediction to spot easy-to-learn cases, and lookup model training on those cases. Tests on two long-tail lookup benchmarks show substantial gains over existing lookup models, especially on extremely long-tail groups. Detailed human analysis maps RPDR's strengths and limits, and a dynamic routing plan is proposed to send queries to specialized lookup modules for further gains.
Catalog record
| CREATOR | Yiming Zhang, Siyue Zhang, Junbo Zhao, Chen Zhao |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17366