Exploring UX Frustration Recognition using Stress Recognition Models and Lightweight Biosignal Features

Abstract

Adaptive user interfaces provide personalized experiences traditionally through customization menus, user profiling, or multi-modal sensing. Stress and cognitive-load-aware AUIs have received significant attention due to their potential to improve accessibility and performance for less tech-savvy users. Particularly, the use of wearable-captured physiological signals allows practical implementations with high usability. However, attention to bad user experience due to heterogeneous skill levels is limited in the literature; studies are commonly found interpreting stress and frustrating user experiences (UX) as equal, and many approaches rely on self-reported or post-hoc analyses. In this study, we investigate the challenges lightweight machine learning models face when inferring UX-induced frustration from wearable physiological signals using stress-trained datasets, highlighting how task specificity achieves up to 19.3% performance gains in frustration recognition tasks, and label granularity makes 4-class classification viable at above-random chance performance (42.3%).

Links

DOI:

code:references.bib

@inproceedings{bib:sergio_CivilDefence_persuasive2026,

author={Sergio, De Le\'{o}n Aguilar and Matsuda, Yuki and Yasumoto, Keiichi},

title={Exploring UX Frustration Recognition using Stress Recognition Models and Lightweight Biosignal Features},

booktitle={21st International Conference on Persuasive Technology (PERSUASIVE '26)},

pages={x--x},

year={2026},

doi={},

url={}

}

https://scrapbox.io/files/698df9da453671c306223ddb.png

Category

Conference

Keywords

Collaborating Organization