Homepage redesign: review notes

Branch redesign-2026. Nothing here is live until this branch is merged into master.

What changed

New look, same URLs. /, /publications/, /resume/, /services/, /awards/ keep their addresses; /news/ is new. Old /about/, /cv/ and /publication/2017-SFL-IJCAI redirect.
Homepage keeps the academic structure: profile card and three-paragraph bio (based on the old About text), venues and recognition, five research areas with representative papers, nine selected publications, news (10 shown, “Show all” for the full history back to 2020), experience and education, honors, academic service.
Publications page: 61 papers grouped by year, filterable by topic (All, Selected, Lead-authored, 8 topics). Filters are linkable, for example /publications/#topic=llm. Each paper has a Figure 1 thumbnail (or a figure chosen from the paper), a one-sentence summary, the venue, awards, and links.
Data files are the single source: _data/publications.yml, news.yml, awards.yml, service.yml, research.yml, topics.yml, site.yml. To add a paper, add one entry to publications.yml; the homepage counts, filters and lists update themselves.
Removed AcademicPages leftovers: sample posts, talks, teaching, portfolio, demo pages and images, files/paper1-3.pdf, the sample _data/authors.yml, the talkmap and markdown generator.
SEO: real site description, Open Graph image, JSON-LD Person, page titles like “Publications Bojian Hou”. The dead google-universal analytics block is switched off.
Experience and education follow the résumé: Meta (Monetization and MRS), Penn (Postdoctoral Fellow), Cornell, 4Paradigm (Research Scientist, 2020 to 2021, new on the site), Nanjing University Ph.D. 2014 to 2020 and B.Sc. 2010 to 2014. The bio now says “postdoctoral fellow” and mentions 4Paradigm.
Quality checks (latest preview build): Jekyll builds cleanly with the GitHub Pages gem set; 125 publication links checked, 0 broken (10 publisher pages refuse bots and are marked “blocked”); no horizontal scrolling at 390 px; Lighthouse mobile: home 99 / 100 / 100 / 100, publications 97 / 100 / 100 / 100 (performance / accessibility / best practices / SEO). Fonts (Geist, SIL Open Font License) are self-hosted in assets/fonts/.
Files: files/CCFAI'19.pdf renamed to files/CCFAI19.pdf (apostrophes break links).

Publications: decisions applied

The list matches Google Scholar (one entry per paper; preprint and published versions merged), plus the CCFAI 2019 paper.
Removed MentalGPT (AMIA 2024 abstract).
Added 10 papers that are on Scholar but were not on the old site, with full author lists copied from the papers: npj Digital Medicine 2026 (CT VLM), arXiv 2026 (TAP-GPT multimodal), AMIA 2025 (Fair multi-modal CCA), IEEE TAI 2025 (ICAFS), Innovation in Aging 2025 (SDoH + EHR + KG), two ICCV 2025 workshop papers, two AIMedHealth 2025 papers, arXiv 2025 (RNA survey).
Title corrections to the published versions: “Determining the Importance of Clinical Modalities …” (AMIA Summit 2025), “… Alzheimer’s Disease Subtype Discovery” (Medical Image Analysis 2024), “Learning Interpretability from RNN with Feature Changing” (CCFAI 2019).
Author corrections from the papers: ISBI 2023 has “Hao Zheng” (not Hao Zhang); the DETR arXiv paper includes “Bohan Chen”.
Code and dataset links added where the paper’s abstract states them (MentalChat16K dataset and code, FREEFORM, SDoHenPKG, DALK, GlaucomaNet, Manifoldron, TAP-GPT, FR-CCA-ADNI).
Links fixed: legacy PMC PDF paths, an ACM link with a personal casa_token, [PDF] labels that pointed to arXiv pages, absolute bojianhou.github.io links.

Please check

Author lists changed to match the published versions (your old page differed):
- ICML 2025: PMLR lists “… Ruochen Jin, Qi Long, Weijie J. Su, Li Shen” (old page had Weijie Su before Qi Long).
- IEEE TAI 2024: the published paper includes Xiaoge Zhang (5th author).
- IEEE TNNLS 2024 (Manifoldron): the published paper includes Zhen Jia and Boce Zhang (Crossref spells “Boce Zhang”; arXiv spells “Boce Zhou”).
- Kunlun keeps “et al.” after six authors (the paper has 28).
Kunlun now links to the ACM DL page (doi.org/10.1145/3770855.3818438) as well as arXiv.
Thumbnails. All figures are cropped from the papers. A few use a figure other than Figure 1 because Figure 1 was a results table or a cohort flowchart; see scripts/site_tools/teaser_overrides.yml.
One-sentence summaries are paraphrased from each abstract (no numbers or claims beyond the abstract). Skim them on /publications/.
Still without a public version: MLMI 2026 (in press; shows “In press”). Fair multi-modal CCA (AMIA 2025) links to PubMed only.
Chinese name. The homepage shows “also published as Bo-Jian Hou · 侯博建” (the old sidebar already showed 侯博建). Remove it if you prefer.
Links that block automated checkers (ACM, OUP, medRxiv DOIs return 403 to bots). They open normally in a browser; the link report marks them “blocked”, not broken.

On Google Scholar itself (optional)

Three AAIC abstracts are still listed (FR-CCA, Neural Additive DCSM, Interpretable DCSM risk groups).
Duplicates to merge: the old-title ICLR 2025 preprint (2 entries, one with a garbled author field), an entry titled only “Uncertainty in Artificial Intelligence”, and preprints of the BCB 2025, TAI 2024, TNNLS 2020, npj 2026 and ESN papers.
Typo: “Prexdiction” in the TAP-GPT multimodal title.
CCFAI 2019 is not on Scholar; add it to make the two lists identical.

Maintenance tools

scripts/site_tools/ (run by .github/workflows/site-tools.yml, never on master):

collect_metadata.py: titles, authors and abstracts from arXiv, Europe PMC and Crossref.
fetch_pdfs.py + extract_teasers.py: download each paper and crop its figure into images/pubs/<id>.webp.
check_links.py: checks every publication link.
screens.mjs: screenshots at desktop and phone widths in light and dark mode, plus filter and news checks.

Trigger from the Actions tab, or push a commit whose message contains [tools] or [preview] to any branch other than master. Results are pushed to the site-tools-output and site-preview-output branches.