brycewang-stanford ICDM Review Process
Understand how ICDM decisions are actually made, so you can write for the process and read the outcome correctly. ICDM's Research Track runs a triple-blind review, has a distinctive short-paper outcome, and traditionally offers no author rebuttal — a very different shape from OpenReview venues with public discussion.
Triple-blind mechanics
- Author identities and referee identities are both concealed. Since 2011, referee names are hidden even from each other during discussion; only the PC Co-Chairs know who is who.
- Author names are disclosed only after ranking and acceptance are finalized — so nothing in the review can be swayed by who you are, for better or worse.
- Practical consequence: the paper is judged purely on the anonymized PDF and its cited anonymized artifact. Write for a reviewer who knows nothing about you.
The outcome space
| Outcome | What it means | Your move |
|---|---|---|
| Accept (regular) | In as a full paper | Camera-ready at full length |
| Accept as Short | Contribution valued, but scoped/evidenced for short length | Compress to the short camera-ready (see icdm-camera-ready) |
| Reject | Not accepted this edition | Revise and route (SDM/CIKM/KDD next window) |
The Accept-as-Short outcome is ICDM's signature: a full submission can be offered acceptance at
short length rather than rejected outright. Plan for it (see icdm-workflow), because it arrives
with the notification and a compressed camera-ready deadline.
The reviewer pool
- ICDM reviewers are data-mining specialists: algorithmic, graph, temporal, and applied-mining researchers. They reward a named mechanism, careful baselines, and defensible discovery, and they punish leaderboard-only novelty and un-checkable claims.
- The Applied Track (single-blind, new in 2026) draws a more application-oriented pool that weights deployment and measured real-world impact.
- Expertise varies within a paper's reviews; the self-contained body (no rebuttal to clarify) is your defense against a reviewer who misreads the setup.
The no-rebuttal posture
- Historically ICDM has no author-response window; whether the current edition adds one is
待核实 (see
icdm-author-response). Assume none when planning. - This makes the submitted PDF and cited artifact your only argument. Every likely reviewer question — baseline fairness, leakage, variance, scale — must be pre-answered on the page.
- A weakness "left for the rebuttal" is a weakness left unaddressed.
Reading a decision packet
1. Find the decision line: Accept / Accept-as-Short / Reject.
2. If Accept-as-Short, read reviews for WHAT to cut vs keep - the mechanism stays.
3. Cluster review concerns: correctness, baselines, novelty, scale, clarity, validity.
4. Separate factual misreadings (fixable in camera-ready framing) from real gaps
(need new experiments -> next venue).
5. Note anything the reviewers could not see because it was missing from the PDF/artifact;
that is your no-rebuttal lesson for next time.
Vignette: reading an Accept-as-Short correctly
A paper comes back "accepted as a short paper": reviewers found the mechanism novel but felt one of three experiments was under-developed. The wrong read is disappointment; the right read is a scoping instruction. The team keeps the mechanism and its strongest experiment in the short body, moves the two weaker experiments and all protocol detail to the cited repository, and ships a tight short paper on time — the contribution is preserved, just at the length reviewers judged it earned.
Output format
[Decision] Accept / Accept-as-Short / Reject
[Regime read] triple-blind (Research) / single-blind (Applied)
[Concern clusters] correctness / baselines / novelty / scale / clarity / validity
[Misreading vs gap] <which concerns are framing-fixable vs need new work>
[No-rebuttal lesson] <what was missing from PDF/artifact that reviewers needed>
[Next move] camera-ready / short-compression / revise-and-route