End-to-End Prior-Guided Graph Attention Network with Differentiable Structural Priors for Interpretable Mobile Interface Layout Scoring

Long Zeng1 Zi Wang2 Wen Zuo3 Jun Hoong Chan4

  1. School of Information Engineering, Chenzhou Vocational Technical College Chenzhou 423000, China zenglong00000@163.com
  2. School of Information Engineering, Chenzhou Vocational Technical College Chenzhou 423000, China wangzi1120@163.com (corresponding author)
  3. School of Fine Arts and Artistic Design, Hunan University of Science and Engineering Yongzhou 425199, China 64012465021@msu.ac.th
  4. School of Computer Science, Peking University Beijing 100871, China junhoong95@stu.pku.edu.cn

Abstract

Automated assessment of mobile-interface aesthetics remains difficult because professional judgments depend on spatial relations among components, whereas image-only models mainly summarize global appearance. This study de velops an interpretable end-to-end scorer for static mobile screenshots. The pro posed prior-guided graph attention network combines a YOLOv10 component de tector, a ResNet-50 visual stream, and a graph stream whose attention is mod ulated by differentiable priors for proximity, alignment, white space, and touch target accessibility. Comparisons with convolutional, graph, and transformer base lines were conducted on app-level splits of the MUD and CLAY datasets. On MUD, the method obtained a weighted root mean square error of 0.387 and a Spearman rank correlation of 0.856, improving the strongest convolutional baseline by 14.3%. Ablation, detector-sensitivity, cross-dataset, and expert blind-test analyses showed that the structural priors improve accuracy, transferability, and diagnostic agreement while retaining end-to-end latency of 35 milliseconds per screen.

Key words

Mobile Interface Aesthetics, Graph Attention Network, Structural Pri ors, Interpretable Regression, Accessibility

Digital Object Identifier (DOI)

https://doi.org/10.2298/CSIS260330052Z