CLAUDE.md — Curry09.github.io

Personal academic homepage built on the acad-homepage Jekyll template (minimal-mistakes based), deployed via GitHub Pages. The whole site is essentially one page: _pages/about.md.

🌏 This site is bilingual (English + 中文) — the #1 rule

The site has a 中 / EN language switch (floating button, top-right). It works entirely client-side: every visible string exists twice — an English copy and a Chinese copy — and CSS shows only one at a time based on a class on <html> (lang-en / lang-zh).

RULE: Any change to visible content MUST be made in BOTH languages. Never add, edit, or remove a piece of visible text in only one language. If you add a news item, a paper, an award, etc. in English, add the matching Chinese (and vice-versa). If you only have one language, ask the user for the other before committing — do not leave a half-translated page.

How the bilingual markup works

1. Inline text (paragraphs, list items, headings) — wrap each language in a span:

<span class="lang-en">English text here</span><span class="lang-zh">中文文本</span>

Keep the two spans adjacent with no space between them (a stray space shows up when switching). Markdown inside a span renders normally: **bold**, [link](url), *italic* all work.

2. Section headings — keep the shared emoji outside the spans, and pin an explicit ID so nav anchors never break:

# 🔥 <span class="lang-en">News</span><span class="lang-zh">最新动态</span> {#news}

The {#news} is required — _data/navigation.yml links to /#news, /#internships, /#publications, /#honors-and-awards, /#education. If you rename or add a section, keep the ID in sync with the nav file.

⚠️ The ID must start with a letter, not a hyphen. kramdown only consumes the {#id} attribute when the ID begins with a letter; an ID like {#-news} is left in the rendered page as literal text (and the heading never gets the ID, so the nav anchor breaks). Use {#news}, never {#-news}.

3. Publications — paper titles and author lists stay English-only (academic convention). Only the bullet descriptions under each paper are bilingual (wrap each bullet’s text in the two spans). Badges, images, links, dates are shared (written once, no spans).

Where each translatable surface lives

The switching machinery (rarely needs changing)

• CSS + pre-paint language script: _includes/head/custom.html
• Toggle button + click handler: _includes/lang-toggle.html (included from _layouts/default.html)
• Default language for first-time visitors: English. The choice is saved in localStorage under site-lang.

Local preview

bundle exec jekyll serve   # then open http://localhost:4000

Click the 中 / EN button and confirm BOTH languages look right before committing.

Citation counts (Semantic Scholar)

Citation counts come from the Semantic Scholar Graph API, in two places:

• Per paper, at the end of each publication’s badge row — a shields.io badge with the Semantic Scholar logo (logo=semanticscholar, -fff?logoColor=000), matching the hand-written Paper/Code/Dataset badges. Built client-side from the fetched count.
• A Total Citations badge in the badge row near the top of about.md, beside Total GitHub Stars and styled to match it (coloured fill, white logo). This one is a shields.io endpoint badge reading s2_data_shieldsio.json straight off the data branch, so it needs no JS. It points at raw.githubusercontent.com like the stars badge does — fine there because shields.io fetches it server-side, not the visitor’s browser.

shields.io treats - and _ as separators inside a /badge/ path segment, so s2Badge() doubles them. An unescaped h-index renders a “404 badge not found” image, which is easy to miss.

Publications are ordered by citation count, descending — not reverse-chronologically. about.md is kept in that order so the first paint is already right, and s2SortPapers() re-sorts on load so the order stays correct as counts drift. A paper Semantic Scholar has no data for sorts last rather than tying with a genuine 0. The sort moves .paper-box nodes between placeholder slots rather than re-appending them to the parent, which would drop them past the Honors and Education sections.

Adding a paper: put the usual arxiv.org/abs/<id> Paper badge on the badge line and append <span class='show_s2_citations' data='<id>'></span>, and drop the block in at its place in the citation ordering. Nothing else to configure — the crawler scrapes the arXiv ids out of the Publications section of about.md, so a paper that isn’t on the Semantic Scholar author profile (co-author disambiguation slips happen) still gets looked up directly.

Notes:

• The author id lives in the workflow (SEMANTIC_SCHOLAR_AUTHOR_ID, currently 2399060433); override it with a repo variable of the same name. An optional SEMANTIC_SCHOLAR_API_KEY secret raises the rate limit but isn’t needed.
• Every paper shows its own count, 0 included. If the branch is missing or the fetch fails the page just renders without citations — no broken text, no dangling |.
• Numbers lag reality: the job runs daily at 08:30 UTC and jsDelivr caches a branch path for up to ~12h. Run the workflow manually from the Actions tab to refresh sooner.
• The data is read through jsDelivr unless citation_stats_use_cdn in _config.yml is set to false; raw.githubusercontent.com is unreachable from mainland China, so the CDN is the default rather than the opt-in.
• Semantic Scholar is the only citation source. The upstream template’s Google Scholar pipeline (google_scholar_crawler, its workflow, and _includes/fetch_google_scholar_stats.html) was removed deliberately — don’t reintroduce it. docs/README-zh.md is upstream template documentation and still describes it. The Google Scholar profile links in about.md and the sidebar are unrelated and stay.
• Run the crawler locally with: cd semantic_scholar_crawler && SEMANTIC_SCHOLAR_AUTHOR_ID=2399060433 python main.py