cikm-review-process

CIKM Review Process

CIKM review runs on EasyChair under double-blind rules, inside the ACM Peer Review

Policy, and — its defining feature — in front of a reviewer pool drawn from three

communities at once. Verified 2026 mechanics (source map, 2026-07-08): submissions

closed in May/June, notification lands August 7, and referees are explicitly barred

from using AI systems to write reviews. Whether 2026 includes an author response

window is unconfirmed in either direction (待核实); plan without assuming one.

The blended-pool effect

A CIKM paper is typically read by people whose default standards differ:

| Reviewer's home lane | What they instinctively grade | Complaint they file most |

|---|---|---|

| Information retrieval | Evaluation design, baselines, metric discipline | "Baselines are stale / significance untested" |

| Data mining | Mechanism novelty, scalability, ablation logic | "Delta over the nearest KDD-line method unclear" |

| KM / databases | Data model, integration cost, system realism | "Would not survive real schema/scale/noise" |

The practical consequence: a paper optimized for one lane can draw its harshest

review from another. Write the submission so each lane finds its own checklist

satisfied — a defensible evaluation, an isolated mechanism, and a credible data

story — or explicitly scope the claim to the lanes it serves. This is also why the

reviewer-pool breadth rewards two-lane papers (see cikm-topic-selection): they

give more of the panel a reason to champion.

Per-track criteria shift

The five tracks are judged against different success definitions. Full Research is

graded on novelty plus evidence depth; Short Research on the sharpness of a single

finding, not breadth; Applied Research on the credibility of deployment evidence

(launch, data release) and transferable lessons; Resource on documentation,

licensing, and likely reuse; Demonstration on what a visitor can actually do with

the prototype. Reviewing a submission against the wrong track's bar — the most

common self-review error — produces both false confidence and false alarm.

What moves a borderline

• A champion, not an average. With three lanes in the room, a decided advocate

who says "my community needs this" outweighs a slightly higher mean score.

• Unanswered lane-specific objections sink. A mining reviewer's unaddressed

scalability question reads as a gap even if both IR reviewers scored high.

• Compliance is upstream of merit. The 2026 desk-reject set — missed reviewer

nomination, undeclared public version, budget or anonymity violations, missing

GenAI disclosure — removes papers before any lane weighs in.

• Chairs calibrate across tracks. Meta-decisions reconcile the lanes; a review

that misread the track's bar can be discounted at that level, which is why a

polite confidential-comments note on track fit (where the form allows one)

is occasionally decisive.

Timeline realism for the live cycle

Between the June close and the August 7 notification there is no author-visible

activity by default. Do not read silence as signal; do not email chairs for status;

do use the window as cikm-workflow Mode A prescribes (artifact readiness,

camera-ready pre-drafting). If a response phase is announced mid-cycle, it will be

short — pre-agree within the team who drafts and who signs off.

Reading a CIKM review packet

When reviews arrive, decode them by lane before reacting:

For each review:
  1. Identify the lane from the vocabulary
     ("nDCG/baselines/collections" → IR; "novelty/ablation/scale" → mining;
      "schema/provenance/real data" → KM-DB)
  2. Separate lane-standard demands (must answer) from
     lane-mismatch complaints (may be a track/framing misread)
  3. Rank objections by whether a chair would treat them as blocking:
     correctness > missing decisive evidence > positioning > polish

Two panel patterns worth recognizing. Split-by-lane scores (one lane high, one

low) usually mean the paper is legible to only part of the panel — a framing

problem more than an evidence problem, fixable at the next venue with

cikm-writing-style. Uniform middling scores usually mean the contribution

is understood and judged thin — an evidence problem no rewrite fixes.

Confidentiality and integrity boundaries

The ACM Peer Review Policy governs both directions. Authors must not attempt

reviewer identification, contact reviewers, or publicize review text with intent

to pressure; reviewers must keep submissions confidential and — a 2026-explicit

rule — must not have AI systems write their reviews. If a review appears

AI-generated or plainly template-pasted, the recourse is a factual, unemotional

note to the program chairs, not a public thread. Chairs can and do recalibrate

around a defective review; they cannot around an author who breached process.

After the decision

Accepted papers move to cikm-camera-ready with an August 20 camera-ready gate —

thirteen days after notification, one of the tightest turnarounds in the family.

Rejected papers should mine the tri-lane reviews for routing information: lane-

specific objections point to the venue whose community wrote them, which is exactly

the input cikm-workflow's fallback ring consumes.

Scale realities

CIKM is one of the largest venues in its family — its proceedings run to many

hundreds of papers across tracks (exact counts per edition: check the ACM DL

record; 2026 statistics 待核实) — which shapes review dynamics in ways small

selective venues do not: reviewer load is high, so the first page carries even more of the

verdict (cikm-writing-style); topical match between paper and reviewer is

looser than at a single-community venue, which is why the abstract's

lane-vocabulary steers bidding; and per-track sub-committees (applied, resource,

demo) apply genuinely different rubrics rather than one program committee's

taste. Acceptance statistics for the current cycle were not verifiable

(待核实) — do not quote a rate to calibrate hopes; calibrate on whether each

lane's blocking question has an answer inside the PDF.

Output format

[Stage] pre-notification / notified-accept / notified-reject / response-window(if any)
[Lane read] IR / mining / KM-DB — likely stance of each on this paper
[Sharpest objection] <the one unanswered question most able to sink it>
[Compliance state] clean / at-risk (which trigger)
[Next move] <single highest-leverage action given the stage>