Reproduce “Layout-aware Webpage Quality Assessment” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Anfeng Cheng, Yiding Liu, Weibin Li, Qian Dong, Shuaiqiang Wang, Zhengjie Huang, Shikun Feng, Zhicong Cheng and 1 more
Researchers at Baidu Inc. developed a GNN-based model for layout-aware webpage quality assessment, processing DOM trees to unify representation across diverse page categories. Deployed in their search engine, this model significantly improved offline metrics, showing a ΔAUC of 24.08%, and enhanced user experience in online A/B tests with a +4.10% increase in user preference for random queries.
Key insights
- DOM trees offer a unified and language-independent foundation for representing webpage layouts, overcoming limitations of previous category-specific feature engineering.
- Graph Attention Networks (GATs) effectively model the intricate, hierarchical structure of DOM trees to infer layout quality, outperforming TreeLSTM and standard GNNs.
- The inclusion of a virtual node allows for comprehensive global context integration, and category-aware modeling significantly improves the assessment's robustness across diverse webpage types.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.