icdm-submission
GitHub用于审计ICDM论文提交准备情况，涵盖研究与应用轨道选择、匿名要求（如三盲审查）、10页篇幅限制及投稿截止日期等关键规则，确保符合会议规范以避免直接拒稿。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill icdm-submission -g -y
SKILL.md
Frontmatter
{
"name": "icdm-submission",
"description": "Use when auditing an ICDM (IEEE International Conference on Data Mining) submission for readiness, covering Research-vs-Applied track choice and its anonymity regime, the triple-blind sweep, the 10-page IEEE all-inclusive cap, the single June deadline, in-paper artifact references, and the desk-level triggers that get an ICDM paper rejected without review."
}
ICDM Submission
Run this audit twice: once when the abstract is registered and once before the full-paper deadline about a week later (2026: abstract May 30, paper June 6, AoE). ICDM runs a single deadline, not KDD-style cycles, so there is no next window this year — the paper you upload in June is the paper that gets reviewed. Reopen the current track call first; Research and Applied now differ on anonymity.
Track and anonymity regime first
- Research Track is triple-blind; the 2026 Applied Track is single-blind; the Blue Sky Track (CCC-sponsored) is a separate visionary lane. Pick the track before formatting, because it decides whether the paper must hide author identity at all.
- One paper, one track: do not submit the same work to Research and Applied at once.
- If the contribution is a deployed system with measured real-world impact, the Applied Track may fit better and its single-blind regime removes the anonymization burden — but confirm the current call's Applied scope before switching.
The triple-blind sweep (Research Track)
ICDM's triple-blind tradition (since 2011) is stricter than double-blind: author and referee identities are hidden, known only to the PC Co-Chairs, and author names are revealed only after acceptance is finalized. Your obligation is to conceal identity throughout the whole paper, not just the title block.
- No author names, affiliations, acknowledgements, funding/grant numbers, or ORCIDs.
- No "in our previous work [X], we..." phrasing that de-anonymizes by self-reference; cite your own prior papers in the third person.
- No identifying system names, internal dataset names, institutional URLs, or repository links that resolve to a named account.
- Scrub figures, plots, PDF metadata, and any embedded file paths.
The 10-page all-inclusive cap
The single hardest ICDM formatting fact: 10 pages in IEEE two-column format, including the bibliography and any appendices. There is no separate reference budget and no separate appendix budget. Over-length papers are rejected without review.
% ICDM IEEE two-column setting (verify against the current track call)
\documentclass[conference]{IEEEtran}
% Everything below counts toward the 10-page cap: body, references, appendix.
% Do NOT shrink margins, fonts, or \baselineskip to reclaim space; IEEE
% proceedings checks and chairs detect a modified template.
% Plan the appendix as a first-class competitor for the same 10 pages as the body.
- Content, references, and appendix together must end by page 10; if it does not fit, cut scope or move detail to an external anonymized repository, never squeeze the template.
- Reference the code/data repository inside the PDF, anonymized: because ICDM traditionally offers no rebuttal, a repository not cited at submission is invisible to reviewers for the whole cycle.
Desk-level and severity table
| Trigger | Track affected | Severity | Fixable after deadline? |
|---|---|---|---|
| Over 10 pages (incl. refs + appendix) | Both | Rejected without review | No |
| Identity leak in body, figure, metadata, or repo | Research | Triple-blind violation | No |
| Modified IEEE template | Both | Desk-level / chair flag | No |
| Same paper in Research and Applied | Both | Rejection of both | No |
| Repository link absent from PDF | Both | Not fatal, but unrepairable without rebuttal | No |
| Wrong track for the contribution type | Both | Hostile review, mis-fit | Only before deadline |
Final-week sequence
- Freeze the track choice and confirm the anonymity regime it implies.
- Rebuild the PDF from a clean
IEEEtrancheckout; measure to the bottom of page 10 with references and appendix included. - Run the triple-blind sweep against the list above (Research Track): names, self-cites, system names, metadata, repo account.
- Export a fresh anonymized repository with no git history and no institutional paths; cite it in the body.
- Register the abstract early in the week; upload the paper with hours, not minutes, of AoE margin.
Vignette: a paper that lost a page to its own bibliography
A graph-mining team writes a tight 9.5-page body and adds a two-page appendix of proofs, forgetting that at ICDM the references (1.5 pages) and the appendix both live inside the 10-page cap. The compiled PDF is 12 pages — an automatic no-review reject. The fix is not cosmetic: they promote one proof sketch to the body, move the full proof and extra tables to the cited anonymized repository, and trim the related-work prose. The lesson: at ICDM, budget the appendix and references against the body from the first outline, because there is no separate allowance to grow into.
Output format
[ICDM readiness] Ready / Needs fixes / Not ready
[Track x anonymity] Research(triple-blind) | Applied(single-blind) | Blue Sky
[Page check] <= 10 pages incl. refs + appendix: pass / fail
[Anonymity] triple-blind sweep clean / leaks found (Research only)
[Artifact visibility] repo cited in PDF: yes / no
[Fix order] <ordered fixes before the AoE deadline>
Version History
- 9f86f09 Current 2026-07-19 15:55
