@inproceedings{balepur-etal-2025-whose,
title = "Whose Boat Does it Float? Improving Personalization in Preference Tuning via Inferred User Personas",
author = "Balepur, Nishant and
Padmakumar, Vishakh and
Yang, Fumeng and
Feng, Shi and
Rudinger, Rachel and
Boyd-Graber, Jordan Lee",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.168/",
doi = "10.18653/v1/2025.acl-long.168",
pages = "3371--3393",
ISBN = "979-8-89176-251-0",
abstract = "LLMs are aligned to follow input instructions by learning which of two responses users prefer for a prompt. However, such preference data do not convey *why* users prefer responses that are chosen or rejected, so LLMs trained on these datasets cannot tailor responses to varied user needs. To surface these parameters of personalization, we apply *abductive reasoning* to preference data, inferring needs and interests of users, i.e., personas, that may prefer either response. We test this idea in two steps: **Persona Inference (PI)**{---}abductively inferring personas of users who prefer chosen or rejected outputs{---}and **Persona Tailoring (PT)**{---}training models to tailor outputs to personas from PI. We show: 1) LLMs infer personas accurately explaining why different users may prefer *both* chosen or rejected outputs; 2) Training on preference data augmented with PI personas via PT boosts personalization and generalizes to supporting user-written personas; and 3) Rejected response personas form harder personalization evaluations, showing PT better aids users with uncommon preferences versus typical alignment methods. We argue for an abductive view of preferences for personalization, asking not only which response is better but when, why, and for whom."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="balepur-etal-2025-whose">
<titleInfo>
<title>Whose Boat Does it Float? Improving Personalization in Preference Tuning via Inferred User Personas</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nishant</namePart>
<namePart type="family">Balepur</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Vishakh</namePart>
<namePart type="family">Padmakumar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fumeng</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shi</namePart>
<namePart type="family">Feng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rachel</namePart>
<namePart type="family">Rudinger</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jordan</namePart>
<namePart type="given">Lee</namePart>
<namePart type="family">Boyd-Graber</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>LLMs are aligned to follow input instructions by learning which of two responses users prefer for a prompt. However, such preference data do not convey *why* users prefer responses that are chosen or rejected, so LLMs trained on these datasets cannot tailor responses to varied user needs. To surface these parameters of personalization, we apply *abductive reasoning* to preference data, inferring needs and interests of users, i.e., personas, that may prefer either response. We test this idea in two steps: **Persona Inference (PI)**—abductively inferring personas of users who prefer chosen or rejected outputs—and **Persona Tailoring (PT)**—training models to tailor outputs to personas from PI. We show: 1) LLMs infer personas accurately explaining why different users may prefer *both* chosen or rejected outputs; 2) Training on preference data augmented with PI personas via PT boosts personalization and generalizes to supporting user-written personas; and 3) Rejected response personas form harder personalization evaluations, showing PT better aids users with uncommon preferences versus typical alignment methods. We argue for an abductive view of preferences for personalization, asking not only which response is better but when, why, and for whom.</abstract>
<identifier type="citekey">balepur-etal-2025-whose</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.168</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.168/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>3371</start>
<end>3393</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Whose Boat Does it Float? Improving Personalization in Preference Tuning via Inferred User Personas
%A Balepur, Nishant
%A Padmakumar, Vishakh
%A Yang, Fumeng
%A Feng, Shi
%A Rudinger, Rachel
%A Boyd-Graber, Jordan Lee
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F balepur-etal-2025-whose
%X LLMs are aligned to follow input instructions by learning which of two responses users prefer for a prompt. However, such preference data do not convey *why* users prefer responses that are chosen or rejected, so LLMs trained on these datasets cannot tailor responses to varied user needs. To surface these parameters of personalization, we apply *abductive reasoning* to preference data, inferring needs and interests of users, i.e., personas, that may prefer either response. We test this idea in two steps: **Persona Inference (PI)**—abductively inferring personas of users who prefer chosen or rejected outputs—and **Persona Tailoring (PT)**—training models to tailor outputs to personas from PI. We show: 1) LLMs infer personas accurately explaining why different users may prefer *both* chosen or rejected outputs; 2) Training on preference data augmented with PI personas via PT boosts personalization and generalizes to supporting user-written personas; and 3) Rejected response personas form harder personalization evaluations, showing PT better aids users with uncommon preferences versus typical alignment methods. We argue for an abductive view of preferences for personalization, asking not only which response is better but when, why, and for whom.
%R 10.18653/v1/2025.acl-long.168
%U https://aclanthology.org/2025.acl-long.168/
%U https://doi.org/10.18653/v1/2025.acl-long.168
%P 3371-3393
Markdown (Informal)
[Whose Boat Does it Float? Improving Personalization in Preference Tuning via Inferred User Personas](https://aclanthology.org/2025.acl-long.168/) (Balepur et al., ACL 2025)
ACL