Skip to main content
Date: 2026-07-17 · For: honest-scholar literature capability, scout mode (sub-spec 2). · Status: verified-source digest; migrates to the plugin’s resources/references/. Using the citation graph to generate research ideas, spot gaps, and find collaborators, anchored on an author’s paper + rival anchors. Identifiers verified at compile time unless flagged [unverified].

​
1. Named methods

  • Forward-citation chasing / snowballing — walk an anchor’s citing set (forward) and reference set (backward), iterate with explicit inclusion criteria; forward surfaces “who built on us / the rivals.” Rigorous protocol: Wohlin, C. (2014), Guidelines for snowballing in systematic literature studies, EASE ‘14, DOI:10.1145/2601248.2601268.
  • Citation context / function / sentiment — classify why a paper cites another (background vs method-use vs result-comparison; supporting vs contrasting). Core primitive for hypothesis-level scouting. Teufel, Siddharthan, Tidhar (2006), Automatic classification of citation function, EMNLP, ACL W06-1613. Cohan et al. (2019), Structural Scaffolds for Citation Intent Classification, NAACL, arXiv:1904.01608 → SciCite (Background/Method/Result-comparison), the scheme the Semantic Scholar API exposes.
  • Co-citation analysis — two papers related if a later paper cites both; clusters reveal a specialty’s intellectual base / research fronts. Small, H. (1973), JASIS 24(4):265–269, DOI:10.1002/asi.4630240406.
  • Bibliographic coupling — two papers related if they share references; works for brand-new papers. Kessler, M.M. (1963), American Documentation 14(1):10–25, DOI:10.1002/asi.5090140103.
  • Research-front detection / gap-finding — the growing tip of a literature. de Solla Price (1965), Science 149(3683):510–515, DOI:10.1126/science.149.3683.510. Operationalized via co-citation clustering + burst detection: Chen (2006) CiteSpace II [unverified]; Kleinberg (2002) KDD burst model [unverified].
  • Main-path analysis — dominant knowledge-flow chain through a citation DAG (weights links, not nodes). Hummon & Doreian (1989), Social Networks 11:39–63.
  • Idea backlog composition (not a single named method) — forward-chase → context-classify → cluster → tag each by idea type (untested extension, contradiction/rival, new domain, transferable technique, methodological gap). See §3.

​
2. Tooling (ranked for a scriptable, reproducible workflow)

Recommendation: OpenAlex backbone + Semantic Scholar for contexts/intents/recommendations. Both free, keyless-or-cheap, JSON, reproducible.

​
3. Process: anchor → ranked ideas + collaborators

  1. Fix the anchor set (own papers + rival anchors); resolve each to a stable ID (DOI/arXiv/OpenAlex/S2).
  2. Pull forward citations per anchor (OpenAlex filter=cites:, S2 /citations); store raw JSON = provenance root.
  3. Enrich each citing paper: year, venue, citation count, abstract, authors+affiliations, citation-context snippet + intent (S2).
  4. Filter/prioritize: intent (Method/Result-comparison > Background; isInfluential), recency (24–36 mo up for fronts), impact (tie-breaker, not primary), coupling proximity.
  5. Cluster by co-citation + bibliographic coupling to expose sub-fronts.
  6. Classify each cluster/paper into an idea type (extension / contradiction / new domain / transferable technique / gap).
  7. Emit a ranked backlog row per idea with mandatory provenance: idea | type | source paper (ID) | citing-context snippet | why-it-matters | est. feasibility.
  8. Collaborators: rank citing-paper authors by frequency × recency × non-overlap with existing coauthors.
  9. Make it standing (Litmaps watch / scheduled OpenAlex query) → monitored feed.
Provenance rule: every generated idea carries the citing paper’s ID + exact citation-context snippet — the auditable link from idea back to who-said-what.

​
4. Level split — same method, two tunings (not two methods)

  • Hypothesis level (narrow): precision-optimized; small set, read full text, citation-context/sentiment dominates (find Contrasting/Result-comparison citations).
  • Paper level (broad): recall-optimized; large set, skim metadata, co-citation clustering + burst/front detection dominates.
→ one pipeline with a level parameter swapping (i) ranking weights, (ii) reading depth, (iii) primary technique. Separate methods would duplicate the graph plumbing.

​
5. Idea-shaping frameworks (back end)

  • Sandberg & Alvesson (2011), Ways of constructing research questions: gap-spotting or problematization?, Organization 18(1):23–44, DOI:10.1177/1350508410372151 — flag mere gap-spotting vs assumption-challenging problematization.
  • Alon (2009), How to choose a good scientific problem, Molecular Cell 36(6):721–723, DOI:10.1016/j.molcel.2009.09.013 — feasibility × interest scoring.
  • Hamming, “You and Your Research” (1986) [unverified] — filter against pure gap-filling.

​
Sources

Small 1973 (asi.4630240406) · Kessler 1963 (asi.5090140103) · Hummon & Doreian 1989 · Teufel et al. 2006 (ACL W06-1613) · Cohan et al. 2019 (arXiv:1904.01608, SciCite) · Price 1965 (science.149.3683.510) · Sandberg & Alvesson 2011 (1350508410372151) · Alon 2009 (S1097-2765(09)00641-8) · S2AG API (arXiv:2301.10140), S2ORC (arXiv:1911.02782) · OpenAlex (arXiv:2205.01833) · Wohlin 2014 (10.1145/2601248.2601268).

​
Flags

[unverified this session] (confident from domain knowledge): CiteSpace (Chen 2006 JASIST 57(3):359–377), Kleinberg 2002 KDD, VOSviewer (van Eck & Waltman 2010), CitNetExplorer (2014), scite.ai API, Inciteful API, Hamming 1986, Wohlin 2014 DOI.