Zichen Ding · Research Portfolio

A responsive academic homepage for Zichen Ding, built with Jekyll and deployed on GitHub Pages. The custom portfolio uses local assets and vanilla JavaScript, with no frontend build step or external font dependency.

Mobile preview

Run locally

Install Ruby and Bundler, then run from this directory:

bundle install
bash run_server.sh

Open localhost:4000. To check the production build:

JEKYLL_ENV=production bundle exec jekyll build

The existing lockfile uses Bundler 2.2.19. If Bundler reports a version mismatch, install that version with gem install bundler -v 2.2.19. Local dependencies under vendor/ and Bundler configuration are ignored by Git.

Update content

• _config.yml: identity, email, social links, site URL, and description.
• _data/publications.json: the publication index. Each entry includes its title, full authors, venue, year, topic, resource links, and optional image / summary. Set selected: true to feature a paper. The homepage owner is highlighted automatically in author lists.
• _data/news.json: news entries, newest first; the first five appear before expanding the archive.
• _includes/homepage.html: biography, research directions, education, honors, service, and contact.
• assets/css/portfolio.css: responsive layout and light / dark themes.
• assets/js/portfolio.js: interactive agent sketch, publication explorer, theme preference, clipboard actions, and navigation.
• images/previews/: compressed publication and portrait previews. Original full-resolution images remain in images/.

Publication topics are gui, science, language, and multimodal. Selected defaults to newest first, using arXiv first-release month and submission sequence. All publications defaults to accepted papers first (including Findings and workshop tracks), with newer venue years first and release order within each year; preprints follow in release order. Manual sort preferences are retained separately for each collection. Text search includes the full bibliography, even when the Selected collection is active. Selecting a research direction also opens the complete matching bibliography. “Copy reference” copies a plain-text reference based on the saved metadata.

Interaction & accessibility

• Click Observe / Reason / Act in the hero to explore the illustrative agent loop. Clicking a stage pauses autoplay; use the play button to resume.
• Search by title, author, venue, or year; press / to focus the search field. Filter by topic and sort by year.
• Expand publication details to read complete author lists and summaries.
• The theme follows the OS preference until the visitor chooses a theme; the choice is stored locally.
• Keyboard focus, a skip link, reduced-motion preferences, mobile navigation, empty search results, no-JavaScript reading, and print styles are supported.
• All academic content is rendered by Jekyll. Essential content and native disclosure elements remain available without JavaScript.

The Google Scholar crawler still publishes gs_data.json and gs_data_shieldsio.json on google-scholar-stats. The page refreshes citations from the existing CDN endpoint, with the raw GitHub endpoint as a fallback. _data/scholar.json holds a verified, dated snapshot for visitors who cannot reach either endpoint; update it from the same Scholar account, retaining its original updated date. The displayed date is the crawler's data date, not the visitor's current date. The workflow uses an isolated Python environment, pins the BibTeX parser API required by scholarly, and updates the statistics branch without force-pushing. It runs daily, on crawler source changes, or manually; ordinary page builds do not trigger extra Scholar requests. Fetches have a three-minute upper bound, validate the author and count before writing results, and date successful retrievals in Asia/Shanghai. GitHub Pages can continue building from main using the repository's existing Pages configuration. Local edits do not publish until committed and pushed.

Credits

Based on AcadHomepage, with a custom portfolio layout. The original template was influenced by Minimal Mistakes and AcademicPages. Original license notices and template files are retained; see LICENSE.

Regression checks

Node.js 18+ is only needed for development tests, not for serving or publishing the site:

npm ci
bundle exec jekyll build
npm test
python3 -m unittest discover -s tests -p 'test_*.py'

The tests exercise search / topic / collection / sort combinations, complete pagination, direct paper links, keyboard navigation, color preferences, reduced motion, denied storage and clipboard access, and Scholar endpoint / snapshot fallbacks. To test the production Pages build, also run JEKYLL_ENV=production bundle exec jekyll build --safe.

Brand assets

The selected dual-77 mark is defined once in _data/brand.json. Run npm ci then npm run build:brand after changing its geometry. This exports the SVG, 16/32px favicons, multi-resolution ICO, touch and Android icons, including a separate maskable icon with safe padding. The page header and footer use the same SVG path.

Company marks: NaiveAI's icon comes from https://naive.ai/icon.svg (retrieved 2026-10-20). Shanghai AI Laboratory uses the existing images/logos/shailab-logo.svg; CSS crops its full lockup to the symbol for legibility beside the company name.

Paper figures and metadata

All 28 publications have covers. _data/figure_sources.json records the source PDF, page, figure number and verified crop for newly extracted or repaired figures. scripts/extract-paper-figures.py reproduces them with PyMuPDF and Pillow; PDF pages are rendered on an opaque white background. Image links progressively open a keyboard-accessible figure dialog, with a direct paper link.

Conference badges follow the CCF 2026 seventh edition. Findings and workshop tracks retain their venue labels but do not inherit the main conference's CCF rank. The Intern-S1-Pro author excerpt preserves source order, includes Zichen Ding, and links to all 177 authors; the full list remains in the details.

DUT's official seal is displayed from the university's VI reference sheet: https://www.dlut.edu.cn/xxgk/dxwh/xxVI.htm (retrieved 2026-10-20).

After changing CSS or JavaScript, run npm run build:assets before the Jekyll build. The generated hashes prevent browsers from retaining an older stylesheet or script after a deployment.