PrefPalette : Personalized Preference Modeling with Latent Attributes
Shuyue Stella Li, Melanie Sclar, Hunter Lang, Ansong Ni, Jacqueline He, Puxin Xu, Andrew Cohen, Chan Young Park, Yulia Tsvetkov, Asli Celikyilmaz · 2025
Read PrefPalette : Personalized Preference Modeling with Latent Attributes by Shuyue Stella Li, Melanie Sclar, Hunter Lang, Ansong Ni, Jacqueline He, Puxin Xu, Andrew Cohen, Chan Young Park, Yulia Tsvetkov, Asli Celikyilmaz (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-2507-13541.txt.
Description
Li and co-authors introduce PrefPalette, a way to model what users like by splitting taste into named traits. Standard preference models treat judgment as a black box. Yet people weigh traits like formality, humor, and cultural fit when they choose. PrefPalette builds training data that isolates one trait at a time. It then learns how social groups weight those traits as they judge. An attention design handles the shifting weights.
The frame draws on multi-attribute decision making from cognitive science. Predictions stay readable, since each trait's role is visible. The method aims to serve groups whose values differ from the average. Aggregate scores, the authors argue, wash out those differences.
Catalog record
| CREATOR | Shuyue Stella Li, Melanie Sclar, Hunter Lang, Ansong Ni, Jacqueline He, Puxin Xu, Andrew Cohen, Chan Young Park, Yulia Tsvetkov, Asli Celikyilmaz |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-13541