brycewang-stanford Web Conference Writing Style
A Web Conference paper is judged by readers who may sit in different disciplines within the same track panel, under a rule that the first 8 pages must carry the entire case. The house style that survives this: lead with the web-native mechanism, quantify the scale early, and write constructs so an adjacent-field expert can check them.
The page-one obligation: web-nativeness
The venue's most predictable review sentence is a version of "unclear why this is a WWW paper." Inoculate on page one by naming the property of the Web your contribution depends on, as a mechanism rather than a backdrop:
| Weak (backdrop) | Strong (mechanism) |
|---|---|
| "We evaluate on web data" | "Anchor text distributes labels across sites, which our method exploits" |
| "Recommendation is important online" | "Feedback loops between ranking and clicks bias naive estimators; we correct for the loop" |
| "Misinformation spreads on platforms" | "Reposting graphs give exposure timestamps that identify the spread model" |
| "We test at web scale" | "At 10^9 edges, quadratic attention is infeasible; our sketch keeps error < ε at linear cost" |
If no such sentence can be written truthfully, the problem is routing, not prose —
go to webconf-topic-selection.
Register for a mixed panel
- Define constructs at first use even when your subcommunity finds them obvious: "engagement (clicks per impression)," "polarization (as measured by ...)." Social-science co-reviewers check construct validity; systems co-reviewers check measurability. One parenthetical serves both.
- Attach every adjective to a number or a mechanism. "Large-scale" means a stated node/edge/user/request count; "efficient" means a complexity bound or a measured cost; "real-world" means a named data source with a collection window.
- Name platforms carefully. During double-blind review, an identifying platform ("our deployment at ") must be masked, but mask with properties, not vagueness: "a shopping platform with ~10^7 daily active users" preserves the evidential weight anonymously.
- Ethics sentences are body text. One or two sentences on consent, ToS
compliance, and aggregation level, in Section 3 where the data is introduced —
not exiled to the appendix (see
webconf-supplementary).
An introduction shape that fits this venue
¶1 The web mechanism and the problem it creates or enables (why the Web)
¶2 Why existing approaches miss it — each named line of prior
work gets a specific failure tied to the mechanism (the gap)
¶3 The contribution, typed explicitly: algorithm / system /
measurement / dataset / theory (what's new)
¶4 Evidence preview with scale attached: datasets, platforms,
windows, headline deltas (why believe)
¶5 Reusability: what transfers beyond this platform/corpus,
plus artifact availability (why care)
Contribution typing (¶3) matters more here than at single-community venues: a measurement paper judged as a methods paper fails, and vice versa. Say which one you wrote.
Compression into 8 self-contained pages
Reviewers may stop at page 8, so compression means re-homing, not shrinking
fonts (acmart tampering is detectable and sanctioned):
- One figure that earns its column: replace the architecture-boxes diagram nobody quotes with the result figure reviewers screenshot.
- Prose-to-table conversions for related work contrasts and dataset statistics.
- Push proof bodies and protocol detail to the appendix, keeping statements and sketches in the body; the appendix is verification, not persuasion.
- Kill roadmap paragraphs ("Section 2 discusses...") — sigconf headers already navigate.
- Cut the second and third examples of every point; web papers over-illustrate.
Sentence-level habits reviewers reward
- Past tense for what you measured, present tense for what the method does, future tense almost never (rebuttal-bait).
- Numbers with uncertainty where runs repeat: "34.2 ± 0.4" or "34.2 (95% CI 33.9-34.5)," and effect direction in words for the cross-disciplinary reader.
- Limitations stated as scope, in the body's final section: "our identification holds for platforms exposing timestamps; feeds without them require..." — this venue reads honest scoping as strength, especially in Responsible-Web-adjacent tracks.
Titles and abstracts in this venue's register
Scan any recent WWW proceedings table of contents and two title patterns dominate: mechanism-first ("Correcting X-Bias in Y with Z") and question-first for measurement papers ("Is X a Y or a Z?" — the Kwak et al. Twitter title is the canonical instance). Patterns to avoid: bare system names with no claim ("WebFooNet"), stacked buzzwords ("Towards Trustworthy LLM-Empowered Graph-Enhanced..."), and titles promising more generality than the evidence holds — the title is the first overclaim reviewers can check. For abstracts, the compression test: a reader should recover mechanism, contribution type, scale of evidence, and one number from the abstract alone. If the abstract survives having every adjective deleted, it is ready; if deleting adjectives leaves nothing, it was decoration all the way down.
Self-edit gate before submission
- Page 1 names the web-native mechanism in one sentence a chair could quote.
- Contribution type declared; every adjective has its number or mechanism.
- A non-specialist in the track's second discipline could restate your main claim after two pages.
- Nothing claim-bearing lives past page 8; no roadmap padding remains.
- Ethics and provenance sentences sit beside the data description.
Finally, keep the register stable across sections: a measurement-precise Section 3 followed by a marketing-voiced Section 6 ("our groundbreaking framework") reads as two authors who never met. The last pre-submission pass should be a single-owner voice edit of the whole 8 pages.
Output format
[Web-native sentence] <quote it, or "missing — routing risk">
[Contribution type] algorithm / system / measurement / dataset / theory
[Register fixes] <constructs undefined, adjectives unattached, platforms named>
[Compression moves] <re-homing applied, pages recovered>
[Remaining risks] <style issues a track panel will still flag>