research-gap-finder

🔍 research-gap-finder

Turn a broad topic into a ranked, citation-backed list of genuine research gaps — ready to become research questions.

This skill operationalizes the field's best gap-identification methods so an AI agent (or a researcher) can run the full pipeline: collect evidence → map the citation landscape → run semantic "what is unstudied?" queries → classify each gap → rank its importance → formulate research questions. Every technique is grounded in the companion catalog in resources.md (100 curated frameworks, tools, databases, and techniques).

Integration reality (read first)

• Open, key-free APIs are available for the scholarly core: Semantic Scholar, Crossref, OpenAlex, Europe PMC, PubMed E-utilities, arXiv. Prefer these for automated queries.
• Most citation-mapping tools (Litmaps, ResearchRabbit, Connected Papers, Inciteful) are browser-based web apps without a public programmatic API. The agent should guide the human user through them step-by-step (or drive them via a browser-automation tool where permitted by their terms of service), rather than pretending to "upload references" via curl.
• Elicit, SciSpace, Scite, Dimensions, Consensus require user accounts (some paid) for full use; the agent must ask the user to provide access or operate them interactively.
• AnswerThis and Aveksana are catalogued as AI research-gap tools; they are interactive web tools — verify their current availability before relying on them.

When to use

• A researcher asks: "what are the research gaps in X?"
• A PhD student needs a defensible gap for a proposal or literature-review chapter.
• A lab wants to scan a field for understudied angles before committing resources.
• An agent must produce an evidence-backed "state of the field + open problems" brief.

⚙️ Operational Workflow

1. Define the search horizon — Draft a broad PICO/PICOS statement (Population, Intervention, Comparison, Outcome, Setting). Record it in Obsidian or Notion as the query template for the whole investigation.
2. Collect recent systematic reviews — Search PubMed, Cochrane Library, and Scopus for the 3–5 most recent systematic reviews or meta-analyses on the topic. Export their citation lists to Zotero.
3. Extract explicit gap cues — For each review, read the "Limitations" and "Future Directions/Research" sections. Summarize each stated gap in a matrix with columns Source, Gap type, Evidence.
4. Map citation networks — Import the reference lists into Litmaps and run the reference-gap method: spot seminal works the reviews failed to cite. Flag each as a potential reference gap.
5. Visualize disconnections — Load the same seed set into ResearchRabbit and Connected Papers. Identify clusters of papers that are never linked ("citation islands"). Record as citation-disconnection gaps.
6. Run semantic gap queries — Use Elicit or Consensus to ask natural-language questions such as "What aspects of X have not been studied in Y population?" Capture the answers plus their supporting citations.
7. Classify gaps with the six-type taxonomy — Tag every collected gap (reference, disconnection, semantic, or author-stated) as evidence, methodological, population, contextual, theoretical, or translational (see Classify & Rank).
8. Assess importance — Score each gap with the five-dimension importance rubric (theoretical, practical, feasibility, novelty, coherence). Enter scores into the matrix.
9. Cross-validate novelty — Check Dimensions, Google Scholar, preprints (arXiv/bioRxiv/medRxiv/SSRN) and grant databases (NIH RePORTER, NSF Awards, EU CORDIS). Downgrade gaps already funded or recently published.
10. Triangulate with external evidence — Check patent databases and industry reports. A gap already patented or industry-validated loses novelty.
11. Prioritize and rank — Sort the matrix by importance score × novelty confirmation. Select the top 3–5 gaps.
12. Formulate research questions — Using PICO/PICOS and the AHRQ framework, rewrite each prioritized gap as a clear, answerable research question.

🤖 AI Tools Playbook

📖 Database Search Strategy

PubMed / PMC — Combine keywords with MeSH terms and systematic-review filters; search "future research"[Text Word] / "further study"[Text Word] in titles/abstracts; snowball via PMID and Cited by.

Scopus — Use citation overview to spot low-citation (understudied) clusters; backward (References) and forward (Cited by) tracking from seminal papers; prioritize Review/Survey types.

Web of Science — Topic search + timespan refinement; Cited References for backward tracking; Related Records (bibliographic coupling) for adjacent understudied areas.

Cochrane Library — Search reviews for "insufficient evidence" / "further research"; cross-reference Joanna Briggs EBP reviews for practice-based gaps.

ERIC (education) — Thesaurus descriptors instead of colloquial terms; filter for meta-analyses; cross-context queries to expose contextual disparities.

PsycINFO (psychology) — Thesaurus + Boolean; search "methodological limitation" in meta-analyses/reviews; find excluded-population gaps via "sample bias" queries.

IEEE Xplore / ACM DL (CS/engineering) — Conference-first: search "future work" / "open problem" in top venues (NeurIPS, CHI, AAAI); survey papers' "remaining challenges".

Google Scholar — "future research" + recent-years filter; author snowballing via Cited by; "knowledge gap" / "contradictory findings" phrase queries.

Universal tactics — forward/backward citation tracking from 3–5 pivotal papers; snowballing (references → search → repeat); dedicate ~30% of search time to scanning Limitations/Future-Directions; log everything into the evidence matrix.

🧭 Classify & Rank Gaps

Six-type knowledge-gap taxonomy

Operationally, a gap must be shown to be a genuine absence (via thorough searching — the four causes of gaps: insufficient/imprecise info, biased info, inconsistent results, or wrong-kind-of-information) rather than an artifact of incomplete searching.

Five-dimension importance rubric (score each 0–3)

Sum the scores (max 15). 13–15 high priority → ready for a proposal; 9–12 medium → needs feasibility work; ≤8 low → re-evaluate relevance. Require at least a 2 in Feasibility to avoid "interesting but impossible".

Evidence matrix (the backbone of the pipeline)

⚠️ Gap identification ≠ importance assessment. Identification ends when the "Gap(s)" cell is non-empty. Importance is judged afterwards with the rubric. Many gaps are genuine but unimportant — never conflate absence of evidence with value of evidence.

🚀 Advanced Mining Strategies

• Patents — Google Patents, USPTO, Espacenet: what's patented but not academically researched.
• Preprints — arXiv, bioRxiv, medRxiv, SSRN: cutting-edge work and emerging gaps before publication.
• Conference proceedings — NeurIPS, CHI, AAAI, ACM, IEEE: mine "future work" sections.
• Theses/dissertations — ProQuest, OATD, EThOS: PhD authors state gaps explicitly.
• Grants — NIH RePORTER, NSF Awards, EU CORDIS: what's funded — and what's not.
• Clinical trials — ClinicalTrials.gov, WHO ICTRP: conditions/interventions with few or no trials.
• Policy & industry — WHO, World Bank, OECD; McKinsey, Gartner, Deloitte: where policy/industry outpace evidence.
• Replication & failures — PLOS ONE, Journal of Articles in Support of the Null Hypothesis: findings that fail to replicate.
• Cross-disciplinary browsing — concepts/methods in adjacent fields not yet applied to yours.
• Bibliometrics — VOSviewer (co-citation/keyword maps), CiteSpace (burst keywords/trends), Bibliometrix (R), Gephi (network visualization), Publish or Perish (citation metrics).

📋 Output: The Gap Report

Document every identified gap with this schema:

• Gap statement — a concise, declarative sentence of what's missing.
• Gap type — six-type taxonomy label.
• Source evidence — the specific papers/reviews/queries that establish the absence.
• Importance assessment — five-dimension rubric scores + total.
• Confidence — High / Medium / Low, based on source quality and search depth.
• Candidate research question — one falsifiable question that would fill the gap.

Honesty & anti-hallucination rules

1. Zero-invented citations — every gap must trace to an identifiable source; otherwise mark it "Exploratory/Hypothetical".
2. Separate identification from importance — always run the rubric; never present a trivial absence as valuable.
3. Label confidence — AI-assisted semantic findings unverified by a review are capped at Medium.
4. No invented trends — if the literature is sparse, say "insufficient literature to determine trends", don't speculate.
5. Absence ≠ algorithm failure — a missing paper in Litmaps/ResearchRabbit may be a search artifact; cross-check with the AHRQ framework before calling it a gap.
6. Verify every citation you output — resolve DOIs/titles via Crossref or a scholarly API before writing them into the report; never emit a citation you cannot resolve.

✅ Quick-Start Checklist

1. Pick 3–5 recent systematic reviews in the area
2. Read every "Limitations" and "Future Directions" section
3. Upload references to Litmaps → find reference gaps
4. Use ResearchRabbit → visualize citation disconnections
5. Use Elicit/Consensus → ask what's unstudied
6. Build the evidence matrix (Study → Method → Population → Findings → Gaps)
7. Classify each gap with the six-type taxonomy
8. Rank gaps with the five-dimension rubric
9. Cross-check grant databases and preprints for novelty
10. Formulate a research question from the most promising gap

📚 The 100-Resource Catalog

The complete, categorized reference list this skill is built on — frameworks, AI tools, databases, strategies, courses, YouTube channels, books, library guides, reference managers, bibliometric tools, communities, key papers, and advanced mining strategies — lives in resources.md.

---

Authored with the repo's AI reasoning team (6 models drafting in parallel, 4 reviewing), and published under the Skill Publishing Standard.