Equity Audit, Trajectory Analysis, and Corpus Chat
Equity and representation audit
scimapR includes built-in tools for auditing geographic, gender, funding, and open-access representation – with explicit confidence reporting and limitation caveats.
Geographic audit
geo <- sm_audit_geographic(corpus)
head(geo)
#> $distribution
#> # A tibble: 20 × 6
#> group count citations pct credit weighted_citations
#> <chr> <int> <int> <dbl> <dbl> <dbl>
#> 1 DE 63 954 9.17 63 954
#> 2 FI 46 708 6.7 46 708
#> 3 AU 44 646 6.4 44 646
#> 4 NO 38 472 5.53 38 472
#> 5 GB 37 544 5.39 37 544
#> 6 IT 37 540 5.39 37 540
#> 7 DK 36 500 5.24 36 500
#> 8 BR 33 533 4.8 33 533
#> 9 CA 33 568 4.8 33 568
#> 10 CH 33 474 4.8 33 474
#> 11 CN 33 541 4.8 33 541
#> 12 NL 33 487 4.8 33 487
#> 13 IL 31 700 4.51 31 700
#> 14 JP 30 464 4.37 30 464
#> 15 FR 29 445 4.22 29 445
#> 16 IN 29 429 4.22 29 429
#> 17 KR 29 442 4.22 29 442
#> 18 US 27 478 3.93 27 478
#> 19 ES 23 341 3.35 23 341
#> 20 SE 23 223 3.35 23 223
#>
#> $gini
#> [1] 0.126
#>
#> $coverage
#> [1] 1
#>
#> $by
#> [1] "country"
#>
#> $weight
#> [1] "count"
#>
#> $method
#> [1] "full"Open access audit
oa <- sm_audit_oa(corpus)
head(oa)
#> $distribution
#> # A tibble: 5 × 4
#> oa_status count pct is_open
#> <chr> <int> <dbl> <lgl>
#> 1 closed 87 43.5 FALSE
#> 2 gold 35 17.5 TRUE
#> 3 green 32 16 TRUE
#> 4 bronze 24 12 TRUE
#> 5 hybrid 22 11 TRUE
#>
#> $pct_open
#> [1] 56.5
#>
#> $coverage
#> [1] 1
#>
#> $n_works
#> [1] 200
Epistemic humility
All audit functions report confidence, method, and caveats. The print methods include explicit limitation blocks. scimapR does not pretend that automated inference is ground truth.
Author trajectory analysis
scimapR can track an author’s career evolution through topic pivots, collaborator turnover, and productivity curves.
# The example corpus includes a trajectory seed author
traj <- sm_author_trajectory(corpus, author_id = "A000000001")
print(traj)
#>
#> ── <sm_trajectory> ─────────────────────────────────────────────────────────────
#> Author: Elena Fischer
#> Author ID: A000000001
#> ORCID: 0000-0003-6689-5331
#> Periods: 5
#>
#> ── Career stages
#> Period 1 (2015-2016): 5 works, h=6 | gene expression, biomarker discovery, drug
#> resistance
#> Period 2 (2016-2018): 6 works, h=9 | colorectal cancer, machine learning,
#> clinical outcomes
#> Period 3 (2018-2020): 6 works, h=11 | single-cell RNA-seq, biomarker discovery,
#> clinical outcomes
#> Period 4 (2020-2022): 10 works, h=13 | biomarker discovery, colorectal cancer,
#> single-cell RNA-seq
#> Period 5 (2022-2024): 13 works, h=14 | immune checkpoint, machine learning,
#> clinical outcomes
#>
#> ── Topic pivots
#> Period 2: score=0.3 (stable)
#> Period 3: score=0.3 (stable)
#> Period 4: score=0.3 (stable)
#> Period 5: score=0 (stable)
#>
#> ── Collaborator turnover
#> Period 2: Jaccard=0.194 (new=14, kept=7, lost=15)
#> Period 3: Jaccard=0.081 (new=16, kept=3, lost=18)
#> Period 4: Jaccard=0.085 (new=28, kept=4, lost=15)
#> Period 5: Jaccard=0.315 (new=22, kept=17, lost=15)
#>
#> ── Emerging collaborators (9)
#> James Ibrahim (since 2024)
#> Carlos Smith (since 2022)
#> Wei Johansson (since 2024)
#> Carlos Patel (since 2024)
#> David Andersson (since 2023)
#> ... and 4 more
#>
#> ── H-index curve
#> 2015-2016:6 -> 2016-2018:9 -> 2018-2020:11 -> 2020-2022:13 -> 2022-2024:14
#>
#> ── Citation acceleration
#> 2015-2016: mean=9.4 (-5.3 vs field)
#> 2016-2018: mean=18.3 (+4 vs field)
#> 2018-2020: mean=17 (+2 vs field)
#> 2020-2022: mean=11.2 (-6.1 vs field)
#> 2022-2024: mean=11.8 (-2.5 vs field)
Author trajectory
LLM-grounded corpus chat
scimapR can use an LLM to answer questions about a corpus, with every claim anchored to actual works via retrieval-constrained citations.
# Requires ellmer package
response <- sm_chat(
corpus,
"What methodological approaches dominate post-2020?",
provider = ellmer::chat_anthropic()
)
print(response)The LLM only sees works that were actually retrieved. Citations are constrained to those retrieved works – no hallucinated references are possible.