A research paper from the School of Economics and Management (SEM) at Beihang University has been accepted and published online by the 39th ACM Symposium on User Interface Software and Technology (ACM UIST 2026), held in Detroit, Michigan, from November 2 to 5. The paper, titled "PrejSense: Non-intrusive Gaze-Informed Prediction of Implicit Prejudice Against Machine for Proactive Interface Adaptation," marks the first time Beihang SEM has published at a top-tier CCF-A conference in human-computer interaction (HCI).

Associate Professor Chen Zhe is the corresponding author, and Beihang University is the primary affiliation. The first author is an outstanding undergraduate from the school. The team presented the work as a video presentation.
Implicit prejudice against machines can critically undermine the effectiveness of human-AI collaboration by limiting trust, cooperation, and system acceptance. Despite its significance, such prejudice is rarely addressed in real-world usage and is often assessed retrospectively through self-report measures or implicit association test, lacking automated and adaptive interface essential for responsive system design.
To address this gap, the team proposes PrejSense, a proof-of-concept adaptive interface loop that uses non-intrusive gaze signals to predict users' prejudice states and adjust interface support accordingly. The system forms a closed loop spanning gaze sensing, prejudice prediction, and interface adaptation.
Interpretable deep learning models predict implicit prejudice from gaze features with 75.8% accuracy. Large language model agents, refined through expert focus groups, then generate personalized interaction strategies. The entire system runs on an ordinary webcam with a latency of just 68.7 milliseconds, enabling unobtrusive, real-time sensing.
In user testing, PrejSense significantly enhances emotional trust and improves the weight of advice (WOA), a standard measure of advice utilization, from 0.690 to 0.842. The results indicate that the system can mitigate algorithm aversion and open a new design space for prejudice-aware adaptive interfaces.
ACM UIST, founded in 1988, is one of the longest-running and most influential conferences in HCI. It is ranked as a CCF-A conference and is widely regarded as a bellwether for interaction technology innovation.
The publication is part of a broader effort by Chen Zhe's group to engage undergraduates in intelligent interaction research. Earlier work from the team on intelligent voice interaction appeared at ACM CHI 2026 and in the International Journal of Human-Computer Studies. All three papers represent the school's first publications at their respective top HCI venues, marking a significant breakthrough in HCI research.
The results also illustrate the school's interdisciplinary approach under the guidance of its management science and engineering discipline, which integrates management with engineering and artificial intelligence. Undergraduates were deeply involved in the full research pipeline, from theoretical modeling and experimental design to deep learning training and system development. This reflects the school's emphasis on strong quantitative foundations, AI proficiency, and the ability to identify and solve complex problems in real-world settings.
Paper link: https://doi.org/10.1145/3830398.3830669
Editor: Lyu Xingyun