⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

PerceptUI: PerceptUI: LLM Agents as Human-Aligned Synthetic Users for UI/UX Evaluation

arXiv 26.06 2026 39.7 method

TLDR

PerceptUI uses persona-conditioned LLM agents to predict user responses for UI/UX evaluation, achieving human-level realism.

Reasoning

The paper introduces a novel two-stage training framework for persona-conditioned UI/UX evaluation, demonstrating strong empirical results across multiple domains. However, it is narrowly focused on UI/UX and does not address world models or dynamics prediction, making the keyword scores zero.

Read-first score

Read-first score 39.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 0.

Recency 6%
100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%
93.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.93585948

Methodology quality 18%
60

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,evaluation

Reproducibility 18%
38

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=dataset

Topical relevance 29%
0

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Citation velocity 12%
0

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontier

Rank sensitivity

Stability: volatile; rank range: 302.

Keyword Scores

world model
0
world simulator
0
generative world model
0
interactive world model
0
video world model
0
world dynamics prediction
0
model-based reinforcement learning world model
0

Deep Analysis

Innovations

  • Persona-conditioned UI/UX evaluation that predicts specific user responses with natural-language rationales
  • Contrastive reflection fine-tuning to distill teacher-generated rationales from human decisions
  • Reflective prompt-evolution step from the model's own failure traces
  • Achieves human-level realism and generalizes to unseen questions and personas

Methodology

PerceptUI is a framework trained in two stages: first, contrastive reflection fine-tuning distills teacher-generated rationales by extracting lessons from human decisions; second, a reflective prompt-evolution step improves the model using its own failure traces. The model is evaluated across multiple domains and datasets.

Key Results

PerceptUI achieves human-level realism in predicting user responses, generalizes to unseen questions and personas, and yields population-level response distributions.

Resources

Paper

Tags