PrefPalette: Personalized Preference Modeling with Latent Attributes

Shuyue Stella Li1,2, Melanie Sclar1,2, Hunter Lang3, Ansong Ni1, Jacqueline He2, Puxin Xu3, Andrew Cohen3, Chan Young Park2, Yulia Tsvetkov2, Asli Celikyilmaz1
1Meta FAIR, 2University of Washington, 3Meta GenAI

Published on Conference on Language Modeling (CoLM) 2025

July 2025

Overview & Motivation

Understanding the relevance to visualization research

Paper Introduction

+

Key Contribution: A framework that decomposes preferences into attribute dimensions and tailors prediction to distinct social community values.

Why This Matters for Visualization

+

Visualization also deals with implicit preferences and needs to consider different user groups and communities.

Research Question

+

How can we model and predict community-specific preferences for better AI personalization?