icse-topic-selection
GitHub辅助判断软件工程论文是否适合ICSE研究赛道，或应路由至FSE、ISSTA等会议及NIER、SEIP等联合轨道。依据贡献类型、证据成熟度及受众进行决策，确保符合ICSE对新颖性、严谨性等要求。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill icse-topic-selection -g -y
SKILL.md
Frontmatter
{
"name": "icse-topic-selection",
"description": "Use when deciding whether a software-engineering project belongs in the ICSE research track or should be routed to FSE, ASE, ISSTA, MSR, ICSME, EMSE\/TSE\/TOSEM, or an ICSE co-located track such as NIER, SEIP, SEIS, or Demonstrations, based on contribution type, evidence maturity, and audience."
}
ICSE Topic Selection
Decide venue before writing a single section. ICSE — the IEEE/ACM International Conference on Software Engineering — is the flagship of the SE research community, which means its reviewer pool expects a contribution to software engineering knowledge, evaluated against the four posted criteria of the research track: novelty, rigor, relevance, and verifiability/transparency (ICSE 2027 CFP wording, checked 2026-07-08). A technically excellent paper whose lesson is about databases, PL semantics, or pure ML will be respected and then rejected as out of scope.
The two-question fit test
- Who changes their behavior if this paper is right? If the answer names software developers, testers, maintainers, reviewers, build engineers, or SE researchers studying them, ICSE is plausible. If the answer names compiler writers, DB implementers, or ML-architecture designers, route elsewhere.
- Would an SE empiricist accept the evidence? ICSE's culture demands evidence proportional to the claim: real subject programs or projects, credible baselines, a threats-to-validity analysis, and (by the open-science policy) an inspectable artifact. A motivating anecdote plus a toy prototype fails the rigor criterion regardless of the idea's quality.
Contribution shapes ICSE rewards
- Technique + tool + empirical evaluation — the classic shape: a new analysis, testing, repair, or synthesis technique evaluated on real programs against real baselines (e.g., automated program repair, whose founding ICSE 2009 paper later won the ten-year Most Influential Paper award).
- Empirical study — quantitative mining, controlled experiments, or qualitative interview/survey work that changes what the community believes about practice.
- Human-factors / developer-experience study — how practitioners actually use (or refuse) tools, with sound qualitative methodology.
- SE for AI and AI for SE — testing ML systems, LLM-based SE techniques; the fastest-growing lane, and also the one with the fastest-staling related work.
- Methodology, benchmark, or dataset contributions when they unlock a research area rather than repackage existing data.
Routing table
| Signal in your project | Better home | Why |
|---|---|---|
| Idea is promising but evaluation is preliminary | ICSE NIER (4 pages + 1 ref page, requires a "Future Plans" section per the 2027 call) | NIER reviews vision, not completed evidence |
| Contribution is a deployed industrial experience | ICSE SEIP | Reviewed on relevance to practice, not research novelty |
| Contribution is the tool packaging or a reusable dataset | ICSE Demonstrations / Data Showcase | Purpose-built venue with interactive review |
| Societal impact of software is the point | ICSE SEIS | Dedicated reviewer pool |
| Testing/analysis is the core, SE framing thin | ISSTA | Deeper testing expertise, same community |
| Automation of a development task, systems flavor | ASE | Sibling flagship, tool-centric |
| Repository mining is the whole method | MSR | Purpose-built co-located conference |
| Maintenance/evolution focus | ICSME | Scope match |
| Study too long or too nuanced for 10 pages | EMSE, TSE, TOSEM | No page ceiling, journal-first return path to ICSE |
| Already published at TSE/TOSEM/EMSE | ICSE Journal-first (2027 window: journal acceptance between Oct 11, 2025 and Oct 10, 2026) | Presentation slot without re-review; secondary studies excluded |
Timing reality for the current cycle
The ICSE 2027 research-track deadline (abstract June 23, submission June 30, 2026, AoE) has already passed as of 2026-07-08, and the 2027 call posts a single submission cycle — a departure from the two-cycle model ICSE ran in 2025 and 2026. So "just aim for the next ICSE window" now means roughly a year of waiting. That makes routing honesty cheaper than it used to be: FSE, ASE, ISSTA, and the journals all have nearer deadlines, and a journal-first acceptance can still put you on the Dublin stage.
Evidence-maturity ladder
Fit is necessary but not sufficient; the same idea belongs at different ICSE doors depending on how far the evidence has come. Grade honestly:
| Maturity rung | You have... | Right door now |
|---|---|---|
| Insight | An observation and an argument, no system or study yet | Workshop or NIER; write the "Future Plans" section the NIER call requires |
| Prototype | A working technique, evaluated on examples you chose | NIER, or hold one cycle and build the real evaluation |
| Evidence | Real subjects, credible baselines, statistics, threats analysis, artifact | Research track — the four criteria are all satisfiable |
| Deployment | Adoption data from practice, weaker research novelty | SEIP, whose reviewers weigh practice lessons over novelty |
| Archive-grade depth | More study than 10 pages can hold | TSE/TOSEM/EMSE first, then journal-first back to the ICSE stage |
The costly mismatch is submitting one rung early: research-track reviewers reject prototypes politely ("promising, but..."), and the year lost to a single-cycle calendar is the real price. One rung of patience is usually one rung of acceptance probability.
Anti-patterns that predict rejection
- Venue shopping downward: a paper rejected from a PL or ML venue, resubmitted to ICSE without re-framing for the SE audience. Reviewers detect the foreign citation graph immediately.
- Solution looking for a problem: no evidence any practitioner or researcher has the pain the tool cures — fails the relevance criterion.
- Benchmark-only novelty: a known technique run on a new dataset with no new question answered.
- Scope camouflage: renaming an ML contribution "SE for AI" without any engineering question. Ask: if the model were swapped, would the SE lesson survive?
Reading the room before committing
Two cheap reconnaissance moves sharpen the verdict. First, scan the last two
editions' accepted-paper titles (ICSE program pages, dblp) for your exact
subarea: three or more recent acceptances means a reviewer pool exists;
zero may mean either an opening or a scope mismatch — the exemplars in
resources/exemplars/library.md help tell which. Second, check where the
papers you cite most were published: if your own bibliography is majority
non-SE venues, ICSE reviewers will experience your paper as a visitor, and
the introduction must do extra naturalization work before the fit test
passes.
Decision procedure
[Audience] who acts differently if the claim is true? -> SE actors? yes/no
[Claim type] technique / empirical / human-factors / benchmark / methodology
[Evidence maturity] complete + real subjects? -> full track
promising + partial? -> NIER or workshop
industrial + deployed? -> SEIP
[Deadline math] next ICSE window vs FSE/ASE/ISSTA/journal calendars
[Verdict] ICSE research track / co-located track / sibling venue + one-line reason
Run this before the writing skills; a wrong venue decision wastes every later step.
When the verdict is ICSE, continue with icse-workflow for the calendar and
icse-writing-style for the paper shape.
Version History
- 9f86f09 Current 2026-07-19 15:59
