Interpretability Analysis of Domain Adapted Dense Retrievers DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation emai; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/18/06
Göksenin Yüksel, Jaap Kamps · 2025
Read Interpretability Analysis of Domain Adapted Dense Retrievers DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation emai; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/18/06 by Göksenin Yüksel, Jaap Kamps (2025) 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-2501-14459.txt.
Description
Goksenin Yuksel and Jaap Kamps study how to explain dense retrievers after they are adapted to a new domain, in work from 2025. Dense retrievers search text with neural codes, but they can falter when the topic shifts.
The authors use a gradient-based method to show which words in a query or document drive the ranking. A new baseline lets them credit both sides. On finance and medical search sets, adapted models leaned more on in-domain terms than plain models did. The work shows this kind of explanation can reveal how adaptation changes model behavior.
Catalog record
| CREATOR | Göksenin Yüksel, Jaap Kamps |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 23 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-14459