A Benchmark and Knowledge-Grounded Framework for Advanced Multimodal Personalization Study

Xia Hu, Honglei Zhuang, Brian Potetz, Alireza Fathi, Babak Samari, Howard Zhou · 2026

Read A Benchmark and Knowledge-Grounded Framework for Advanced Multimodal Personalization Study by Xia Hu, Honglei Zhuang, Brian Potetz, Alireza Fathi, Babak Samari, Howard Zhou (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-2602-19001.txt.

Description

Personal assistants are expected to reason over multi-modal life history, from recognising people to summing up patterns. Yet existing benchmarks mostly test concept recognition. Xia Hu, with six co-authors, introduces Life-Bench in 2026. It is a fully synthetic, human-verified multi-modal benchmark of over eleven thousand question-answer pairs across ten tasks.

Tasks are organised by evidence scope, from concept identification to event understanding to combined reasoning. Photo-centric user histories match real user accounts distributionally. The linked structure of user data invites graph solutions. The authors propose LifeGraph, a user knowledge graph frame with structured lookup and on-demand access to source visual evidence. Evaluation of four lookup paradigms shows accuracy falls sharply with evidence scope, below four-tenths on combined tasks, and no paradigm dominates across categories. Personalisation over multi-modal history remains an open challenge.

Catalog record

CREATORXia Hu, Honglei Zhuang, Brian Potetz, Alireza Fathi, Babak Samari, Howard Zhou
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-19001

Browse the full library · About the Tower of Babel