cikm-topic-selection

CIKM Topic Selection

CIKM is the room where three communities meet: information retrieval, data mining, and

knowledge management with its database roots. Its sponsorship tells the story — ACM

SIGIR (since 1993) plus SIGWEB (since 2006) — and its 2026 self-description spans

information access, knowledge representation, data integration, and personalization

(source map, checked 2026-07-08). Fit questions here are therefore different from

single-community venues: the issue is rarely "is this in scope" and usually "is the

breadth an advantage for this particular paper."

The tri-community fit question

Ask which communities the paper genuinely speaks to, then count:

• Two or three lanes — a retrieval model built on a knowledge graph, a mining

method whose output feeds a search or QA system, an entity-resolution pipeline with

both database and IR evaluation. This is the CIKM sweet spot: the blended reviewer

pool is an asset, because no single-community venue gives all lanes credit.

• Exactly one lane, done deeply — pure ranking-model work, pure pattern-mining

theory, pure query optimization. CIKM accepts such papers, but the specialist venue

(SIGIR, KDD, SIGMOD/VLDB) usually offers sharper reviewers and more career signal.

Choose CIKM deliberately, e.g., for the calendar, not by default.

• Zero lanes — contributions about model architecture, vision, or generic NLP with

an incidental retrieval experiment. Reframe or reroute; scope-mismatch reads fast

at a venue whose name states its scope.

Honest calendar positioning

CIKM's May deadline and November meeting make it the information family's **fall

harvest**: work that missed or matured past SIGIR's January gate, KDD's February

cycle-2 gate, or that wants a decision before WSDM's August gate. Using CIKM as the

next stop after a sibling rejection is normal community practice — but only after the

prior reviews are actually absorbed. A resubmission with untouched weaknesses will

meet a reviewer pool that overlaps heavily with the one that wrote those reviews.

Sibling contrast table

| Neighbor | Its center of gravity | Send the paper there instead when... |

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

| SIGIR | Retrieval and its evaluation culture | Ranking quality/IR evaluation is the entire contribution |

| KDD | Mining at scale; deployment evidence (ADS) | The story is data-regime scale or a production launch |

| WSDM | Web/social interaction data, small selective single track | The object of study is specifically web or social-web behavior |

| TheWebConf | The web as an artifact and ecosystem | Web standards, platforms, or web-scale measurement lead |

| SIGMOD/VLDB | Data systems internals | The contribution is storage, indexing internals, or query processing |

| ISWC / KR | Semantic-web formalisms; formal reasoning | Ontology languages or logic, not applied knowledge systems |

The reverse routes matter too: knowledge-graph construction and reasoning for

applications, heterogeneous-data integration, entity linking, and enterprise search —

work each specialist venue treats as adjacent — is core CIKM territory.

Five-track fork (2026 lineup)

CIKM 2026 fields Full Research (≤10 pp incl. appendices), Short Research (≤4 pp),

Applied Research (≤7 pp, requires practical-application evidence such as a system

launch or data release), Resource (≤4 pp), and Demonstration (≤4 pp) tracks. Pick by

the strongest honest artifact: a rounded method study → Full; one crisp finding →

Short; a deployed system with launch evidence → Applied; a dataset/tool others will

adopt → Resource; something a person can touch at a booth → Demo. The concurrent

submission ban applies across venues, so the same contribution cannot hedge.

Decision walk-through

A team built a temporal knowledge-graph completion model and wired it into an

enterprise document-search assistant. As pure KG-embedding work it competes at scale

venues against heavy theory; as pure search work the IR delta is modest. Framed as

one system — completion quality and its measured effect on retrieval answers — it

becomes a two-lane CIKM paper, with the deployment variant a plausible Applied

Research submission if launch metrics exist. That reframing decision is the routing

decision.

Scope vocabulary sanity check

The venue's name is a scope test you can run on your own abstract. Underline every

noun phrase; a CIKM-shaped abstract naturally contains the venue's working

vocabulary — retrieval, ranking, mining, knowledge graph, entities, integration,

indexing, recommendation, query, corpus, metadata, provenance — because the paper is

about information and knowledge objects. If the underlined terms are instead

"architecture, layers, loss, benchmark, SOTA," the paper is currently written as a

model paper that happens to touch information data, and either the framing or the

venue should change. This check costs five minutes and predicts the most common

CIKM review complaint: "unclear what this offers the information-management

community specifically."

Prestige and career honesty

Advisors and hiring committees read CIKM correctly as a large, credible,

first-tier-adjacent ACM venue — bigger-tent than SIGIR/KDD/WSDM, with more room for

boundary work and applied contributions, and with genuine landmark papers in its

back catalog (the exemplars library holds six, TAGME through BERT4Rec). Honest

implications:

• Choosing CIKM first is a positive strategy for boundary papers, resource

papers, and applied systems — the audiences those papers need actually attend.

• Choosing CIKM second, after a specialist-venue rejection, is routine and

carries no stigma — but the paper must be revised, not merely reformatted,

because reviewer pools overlap.

• Choosing CIKM to avoid a specialist community's standards backfires: that

community's reviewers are in the CIKM pool too, applying the same standards

with less patience.

Timing as a tiebreaker

When fit is genuinely even between CIKM and a specialist sibling, let the calendar

decide like a project manager: a result solid in March comfortably makes CIKM's

May gate; the same result would wait until KDD's next cycle or SIGIR's January.

Six months of shelf life is a real cost in fast-moving areas — LLM-adjacent

retrieval and KG work can be scooped in one arXiv season. Conversely, a result

that needs one more quarter of experiments should not be rushed into May; the

fall slot returns every year.

Output format

[Lane count] IR / mining / KM-DB lanes the contribution truly occupies
[CIKM verdict] sweet spot / acceptable-but-specialist-venue-stronger / out of scope
[Track] full / short / applied / resource / demo (with the honest-artifact reason)
[Alternative venue] <best single-community target and why it loses or wins>
[Framing action] <what to merge, split, or rename before writing>