brycewang-stanford Web Conference Topic Selection
Two decisions, in order: is the work web-native, and which track's reviewer pool should judge it. Both are made by the authors, both are effectively irreversible at the deadline, and the second is as consequential as the first because the venue reviews inside tracks.
Decision 1: the web-nativeness test
Ask: if the Web's specific structure disappeared, would this contribution still make sense? Web-native work depends on at least one of: open hypertext and link structure; platform mechanics and incentives; live, adversarial, user-generated content; web-scale heterogeneity; or the socio-technical coupling of users and algorithms. A generic model that merely evaluates on a web dataset fails the test — the dataset is swappable, so an ML venue's pool serves it better.
Web-native? -> route
Contribution needs link/platform/user -> Web Conference candidate; go to
structure to exist Decision 2
Contribution is a general method; web -> NeurIPS/ICML/ICLR or KDD, cite
data is one evaluation among many web results as evidence
Contribution is about people/society, -> ICWSM or WebSci if measurement/
computation is instrumental interdisciplinarity dominates
Contribution is retrieval effectiveness -> SIGIR first; Web Conference if the
per se open-web setting changes the problem
Contribution is mining methodology with -> WSDM (methods-first) or KDD
modest web specificity (mining/deployment-first)
Contribution is a deployed web system's -> Web Conference Industry track, not
practice lessons the research tracks
Decision 2: track fit (2026 lineup)
The 2026 research tracks, with the question each pool is primed to ask:
| Track (2026) | The pool's primary question |
|---|---|
| Economics, online markets and human computation | Are incentives/welfare modeled, not just predicted? |
| Graph algorithms and modeling for the Web | Does the graph method respect web-graph properties? |
| Responsible Web | Are harms, fairness, or governance the contribution? |
| Search and retrieval-augmented AI | What beats strong current retrieval/RAG baselines? |
| Security and privacy | Is there a real threat model and adversary? |
| Semantics and knowledge | Does it advance KGs/ontologies/structured data on the web? |
| Social networks and social media | Is the social measurement or model construct-valid? |
| Systems and infrastructure (Web, mobile, WoT) | Does it hold under realistic load/deployment? |
| User modeling, personalization and recommendation | Is the personalization gain real and leak-free? |
| Web mining and content analysis | Is the extracted signal novel and robust at scale? |
Tie-breaks: pick the track whose evidence you actually have, not whose name flatters the abstract. A RAG-for-recommendation paper with strong offline ranking tables but no retrieval-baseline sweep survives better in "User modeling ..." than in "Search and retrieval-augmented AI." Track names are re-cut most editions — confirm the current list before deciding, and remember the cap: at most 7 submissions per author across all research tracks in 2026.
Lane selection inside the venue
- Full research paper (8+refs+appendix ≤ 12 pages): a complete evidence program. The default.
- Short paper (4 pages incl. references in 2026; main proceedings): one sharp idea with focused evidence; the later deadline (+6 weeks in 2026) is a schedule fact, not a quality discount.
- Web4Good (2026 special track; main proceedings): work whose center is measurable societal benefit; do not spin ordinary work into it.
- Industry track: deployment truth over methodological novelty; different platform (OpenReview in 2026) and reviewer expectations.
- Workshops (companion proceedings): early-stage or community-building work.
The two chronic misroutes
- The dressed-up ML paper: a general architecture with WWW formatting and one web dataset. Track panels detect it with the swappability question, and the review that follows ("better suited to a specialized venue") wastes a year. Route it to the ML flagship and cite the web experiment as evidence there.
- The homeless interdisciplinary paper: strong social-science finding, thin computational novelty, submitted to a methods-leaning track. It needed "Social networks and social media" (measurement-tolerant) or ICWSM. The fix is pool choice, not more models.
Boundary cases of the current era
- LLM/RAG papers. The venue added "Search and retrieval-augmented AI" as a 2026 track, so LLM work is welcome — when the Web is load-bearing: retrieval over live/adversarial corpora, LLM-generated content polluting web ecosystems, agents navigating real sites. An LLM fine-tuning method evaluated on static QA benchmarks remains an ML-venue paper wearing a web costume.
- Dataset and benchmark contributions. The 2026 lineup had no dedicated resource track (unlike some earlier editions and unlike KDD's separate datasets-and-benchmarks track); a dataset paper competes inside a topical track on the strength of its measurement or enabling analysis. Check the current edition's calls before assuming either way.
- Web3/blockchain, WoT, and mobile. In 2026 these lived inside "Systems and infrastructure for Web, mobile, and WoT" and the economics track rather than as standalone tracks — evidence expectations follow the host track, not the subfield's own conferences.
- Ethics-centered work. If harms analysis is the contribution, "Responsible Web" is the pool primed for it; if harms are one section of a methods paper, stay in the methods track and let the ethics paragraph do its job.
Commitment checklist
- One-sentence web-native mechanism written (it becomes page 1 —
webconf-writing-style). - Track chosen by evidence held; second-choice track named in case the lineup changes at the portal.
- Lane chosen (full/short/Web4Good/industry/workshop) with the calendar
(
webconf-workflow) in view. - Per-author submission counts across the team checked against the cap.
- The nearest sibling-venue alternative named, so the team knows its plan B.
Output format
[Web-native] yes: <mechanism> / no: <reroute target>
[Track] <primary> (evidence basis); fallback <secondary>
[Lane] full / short / Web4Good / industry / workshop
[Misroute risks] <dressed-up ML / homeless interdisciplinary / track mismatch>
[Plan B] <sibling venue + what reframing it would need>