Toward Personalized Social Robots for Child Well-being: Data Requirement Principles from a Recommender-System Perspective

Jul 16, 2026

Upload by Liam

Social robots increasingly support children’s well-being in clinical settings, yet effective help must adapt to each child over time. Personalization can be modeled as a recommender problem that selects the best robot action for a given child, relying on profiling, ranking, and responsible computing. Adoption is limited by data difficulty: user state is non-stationary, session signals are weak and implicit, cross-session identity is disrupted by consent and anonymization, and observational data is off-policy biased. To address this, four data principles are proposed with concrete collection guidelines.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages5

Reading Time13 min

Introduction

Data Collection

Integrated Personalization

Robot-Action Level Data Profile

Record Stream

Accepted as a Late Breaking Report (LBR) at IEEE RO-MAN 2026 Toward Personalized Social Robots for Child Well-being: Data Requirement Principles from a Recommender-System Perspective Jin Huang 1 ,∗ , Eric Nichols2 ,∗ , Fethiye Irmak Do˘gan 1 , and Hatice Gunes 1 arXiv :2607 .05 1 10v 1 [ cs .HC] 6 Jul 2026 Abstract—Social robots are increasingly deployed in clinical settings to support the well-being of children, where effective support must be personalized to each child. Personalization, choosing the robot action best suited to each child, can be framed as a recommendation problem, and a recently proposed recommender-system framework for social robots offers a principled approach through user profiling, ranking, and responsible computing. Instantiating it, however, is blocked not by the model but by the data, which is hard to gather. A child’s state shifts within and across visits, so no fixed description of the user holds. Within a session, the few signals of whether the robot’s actions helped are weak and indirect. Across sessions, children are rarely seen more than once, and anonymization breaks the identity needed to link visits. Because care cannot be randomized, existing data is observational, biased toward whatever was already done. Each is a familiar recommendersystem problem, and we propose four data principles in response: an integrated profile, effectiveness signals, linkable coverage, and an exposure record logged at collection time. We identify which of these principles each capability requires, and frame them as concrete guidelines for data collection. I. INTRODUCTION Ensuring the well-being of children is a priority shared by parents, educators, and clinicians, and it depends not only on a child’s physical care but on how the people around them respond to their needs. This is most acute in clinical settings, where illness and medical procedures place children under stress. Social robots are increasingly brought in to offer companionship and emotional reassurance, and to help children understand what is happening to them [2]–[4] . However, children vary widely in developmental stage, anxiety, prior knowledge, and language, so effective support must be personalized to each child [5], [6] . ∗These authors contributed equally and are co-first authors. 1J. Huang, F. I. Do˘gan, and H. Gunes are with the University of Cambridge, Cambridge, United Kingdom {jh2642, fid21, [hg410}@cam.ac.uk](hg410}@cam.ac.uk) 2E. Nichols is with Honda Research Institute Japan, Wako, Japan [e.nichols@jp.honda-ri.com](e.nichols@jp.honda-ri.com) The work of J. Huang was supported by the EU’s Horizon Europe research and innovation programme under the Marie Skłodowska-Curie Actions Postdoctoral Fellowships (European Fellowship) 2024, grant agreement no. 101203728 — SOCIALADAPT — HORIZON-MSCA-2024-PF- 01. The works of F. I. Do˘gan & H. Gunes have been supported by CHANSE & NORFACE through the MICRO project, funded by ESRC/UKRI grant ref. UKRI572 . Views and opinions expressed are those of the author(s) only and do not necessarily reflect those of the funding bodies. Neither the EU nor the granting authorities can be held responsible. Contributions: Conceptualization: JH, EN. Funding acquisition: JH, HG. Methodology: JH, EN. Writing, Review & Editing: JH, EN, FID, HG. Supervision & project administration: JH, HG. This need is concrete in real-world applications. For instance, at Hospital Virgen del Rocío in Seville, Spain, the social robot Haru [1] provides companionship and conducts anxiety screening for children undergoing cancer treatment [7] . Many of these children, such as those with leukemia, return repeatedly over months of care, and their anxiety, understanding, and emotional state differ from one child to the next and shift from visit to visit. To effectively support them, the robot must adapt both to the individual child and to how that child changes over time. Personalization is, therefore, the capability these

FAQs

Why is personalization necessary for social robots supporting child well-being?

Children differ in developmental stage, anxiety, understanding, and language, and their states shift across interactions and visits. Personalized support is required to choose robot actions suited to each child at the right time.

What makes the required data hard to gather?

The child’s state changes within and across visits, signals within a session are few and mostly indirect, cross-session histories are rarely linkable due to consent and anonymization, and because care cannot be randomized the data is observational and biased.

What four data principles does the paper propose to enable personalization?

It proposes an integrated profile, effectiveness signals for implicit feedback, linkable coverage for connecting histories across visits, and an exposure record logged at collection time to correct for non-randomized care.

Jul 16, 2026

Upload by Liam