Web Conference Review Process

The Web Conference reviews papers inside topical tracks, and that single design choice explains most of what authors experience: who reviews the paper, what evidence bar applies, and why two similar papers can meet very different fates in the same year. This skill models the pipeline as verified for the 2026 edition; per-track reviewer counts, scoring scales, and meta-review mechanics were not published in verifiable sources and are marked 待核实 where they matter.

Pipeline with 2026 anchors

Between December 1 and January 13 there is nothing an author can legitimately do. Emailing chairs with clarifications or new results during deliberation is at best ignored and at worst remembered.

What track scoping means for your reviews

Each of the ten research tracks (from "Graph algorithms and modeling for the Web" to "Responsible Web" to "Systems and infrastructure for Web, mobile, and WoT") recruits its own expert pool via its track chairs. Consequences worth planning for:

• The bar is track-local. "Search and retrieval-augmented AI" reviewers expect ablations against current retrieval baselines; "Economics, online markets and human computation" reviewers expect incentive or welfare reasoning. The same recommendation-systems paper is a methods paper in one track and a measurement paper in another.
• Misrouting is a review-quality event, not just a fit event. A misrouted paper is reviewed by people missing the context to appreciate it, and there is no verified mechanism for authors to move a paper between tracks after the deadline.
• Interdisciplinarity cuts both ways. The venue's identity is computing plus social science plus economics; a track can staff a mixed panel, so a paper should be legible to a smart adjacent reader — define constructs, don't lean on subcommunity jargon for load-bearing claims.

The venue's recurring decision logic

Reviews at this venue orbit four recurring questions; strong submissions answer all four in the first two pages (see webconf-writing-style):

1. Is the problem of the Web — would it exist without platform structure, open hypertext, adversarial content, or live user behavior?
2. Is the evidence at a scale and realism the claim requires?
3. Is the method or finding reusable beyond the studied platform or dataset?
4. Are harms, ethics, and data provenance handled at the standard the "Responsible Web" era expects, whatever the track?

A paper failing question 1 collects the venue's signature phrase — "better suited to a specialized venue" — regardless of technical quality.

Confidentiality and integrity

Double-blind runs in both directions: authors must not attempt reviewer identification or contact, and reviewer materials are confidential permanently, not just until notification. The 2026 CFP's LLM clause also has a review-side face: policies on machine-generated reviews are set per edition and were not verifiable for 2026 — if you review, check the current reviewer instructions before using any AI assistance, and as an author do not assume LLM-written reviews are contestable on that basis alone.

Translating review language

Track panels at this venue have recognizable dialect; decode before reacting:

• "The technical contribution is limited" from an in-track expert usually means "the delta over the sibling-circuit baseline I have in mind is small" — find and name that baseline in the rebuttal rather than defending novelty abstractly.
• "The paper would benefit from evaluation on real-world data" almost always means provenance, not size: the datasets used look synthetic, stale, or undocumented (see webconf-experiments on freshness as evidence).
• "The societal implications deserve more discussion" is rarely a request for a philosophy section; it asks for the two concrete sentences on harms, consent, and aggregation that belong beside the data description.
• "Better suited to a specialized venue" is the web-fit failure; it is nearly unrecoverable in rebuttal unless the web-native mechanism genuinely exists in the paper and merely went unstated — in which case quote it by page and line.
• "I am not an expert in this specific area" in a low-confidence review is routing information for you: the track pool did not contain your reader, which is evidence about track choice for the resubmission.

Reading the outcome

Accept (research tracks)  -> main proceedings, 12-page/8-content budget,
                             registration + presentation obligations attach
Reject                    -> full re-submission somewhere; reviews travel with
                             you informally (reviewer pools overlap with WSDM/
                             CIKM/SIGIR), so fix rather than re-roll
No verified middle states -> the 2026 sources showed no conditional-accept or
                             revise-and-resubmit lane for research tracks (待核实
                             before assuming one exists in your cycle)

After a rejection, the highest-information artifact is the agreement structure across reviews: three reviewers converging on evidence-scale doubts is a webconf-experiments problem; divergent reviews with one champion is often a framing problem (webconf-writing-style), and a routing error is a webconf-topic-selection problem for the next cycle.

One structural note for expectation-setting: the notification is a single decision on a single date (January 13 in the 2026 cycle) — the sources showed no staged early-reject or two-phase review for the research tracks, unlike some ML flagships. Plan emotionally and logistically for one verdict, fifteen weeks after the abstract.

Output format

[Stage] <where the paper sits in the pipeline>
[Track lens] <what this track's pool predictably asks>
[Four questions] web-native? / evidence scale? / reusability? / responsibility?
[Actionable now] <the only levers currently open to the authors>
[待核实 items] <per-cycle mechanics to confirm on official pages>