Waseda computational social scientist stress-testing LLM and agent-based simulators against human data
2
papers here
2
reviewed by Evolve
2
first-author
1
corresponding
—
works · OpenAlex
0
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
This axis measures publication volume and productivity over time. Coverage is incomplete — h- and i10-index read as placeholders (0), with per-year counts only for 2024 (2) and 2026 (4) — so no output score is computed yet.
Impact—
This axis would measure citation uptake and scholarly influence — citations, h-index, i10-index. This OpenAlex record shows placeholder zeros with no usable citation coverage, so impact is not scored and no inference about influence should be drawn.
Focus0/100
Focus scores 0, the strongest dispersion signal in this profile. Evolve's field assignment splits exactly 50/50 between two neighbouring subfields — agent-based modeling with simulation-based inference and LLM-based social simulation — while OpenAlex topics span five clusters: complex network analysis techniques (3), multi-agent systems and negotiation (2), plus single-count work on brand-consumption behavior, AI ethics, and bioinformatic/genomic networks. The two catalog papers, however, sit squarely inside generative social simulation.
Collaboration36/100
Collaboration scores 36, a modest footprint. The recent window lists 5 distinct coauthors, with Chao Li, Xu Wang, and Masayuki Goto recurring (Xu Wang appears twice in the source list, likely a dedup artifact). Both catalog papers are first-authored and one is also corresponding, indicating a lead-author working mode rather than large consortium participation. Small-team, methodology-driven collaboration fits the validation-focused agenda.
Activity57/100
Activity scores 57, mid-range. OpenAlex records 6 works in the recent three-year window, with a clear ramp: 2 works in 2024 against 4 in 2026 — both catalog-reviewed papers fall in 2026. The record lacks 2025 counts, so the trend between those endpoints is unobserved. The current pace points to an active, preprint-first phase centered on generative social simulation.
Reproducibility55/100
Reproducibility scores 55, averaged over the 2 papers Evolve assessed for reproducibility, both rated medium readiness. The practices behind that rating are visible in the papers themselves: one preregistered study replaying human deliberation, and an SBI pipeline with posterior-predictive checking, repair, and a held-out audit. The medium rating suggests solid methodological transparency with room for stronger artifact release.
Research summary
Tengfei Shao is a computational social scientist at Waseda University, Japan, working on validating generative social simulators — both agent-based models and large language model (LLM) agent populations — against empirical human data. The Evolve catalog carries two 2026 first-authored arXiv papers. The first brings simulation-based inference (SBI) into social-simulator validation: amortized neural posterior estimation, posterior-predictive checking, repair, and a held-out audit that tests rather than presumes adequacy against emergent network structure; Evolve's review calls the contribution rigorous and honest while noting the tested simulator remains inadequate. The second, first- and corresponding-authored, is a preregistered replay of human Wason-selection-task deliberation in which belief-anchored LLM groups overstate consensus relative to the human groups by 34–44 percentage points. All bibliometrics here come from OpenAlex: 6 works in the recent three-year window (2 in 2024, 4 in 2026), 5 distinct coauthors, and a broad topic spread spanning complex network analysis, multi-agent systems and negotiation, AI ethics, brand-consumption behavior, and bioinformatic networks. Evolve's field assignment splits evenly between agent-based modeling with simulation-based inference and LLM-based social simulation; the output and impact dimensions lack sufficient data.
Collaboration pattern
The recent three-year OpenAlex window lists 5 distinct coauthors, with Chao Li, Xu Wang, and Masayuki Goto recurring; both catalog papers are first-authored and one adds corresponding authorship, pointing to a small-team, lead-author collaboration style. Notably, the researcher's own name also appears in the frequent-coauthor list and Xu Wang is listed twice — likely name-resolution artifacts to keep in mind when reading the network.
Research keywords
Highlights
Bringing simulation-based inference to social-simulator validation
LLM groups overstate consensus when replaying human deliberation
Timeline
- 2024The OpenAlex record opens with 2 works in 2024; no papers from this year are yet in the Evolve reviewed catalog.
- 2026A first-author arXiv paper introduces SBI-based adequacy testing, repair, and a held-out audit for generative social simulators (Evolve-reviewed; medium reproducibility readiness).
- 2026A first- and corresponding-author arXiv paper reports that belief-anchored LLM groups overstate consensus by 34–44 percentage points when replaying human Wason-task deliberation (preregistered; Evolve-reviewed).
- 2026OpenAlex logs 4 works in 2026 — double the 2024 count — bringing the recent three-year total to 6.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- All bibliometrics come from a single OpenAlex record; h-index and i10-index read as placeholder 0, and the output and impact dimensions are marked missing rather than truly zero.
- The profile is name-resolved from public bibliographic data (no linked ORCID); "Tengfei Shao" is a common name, and the researcher's own name appearing among frequent coauthors suggests residual identity noise.
- Qualitative assessments rest on only 2 Evolve-reviewed papers, both 2026 arXiv preprints with medium reproducibility readiness; peer-reviewed versions are not reflected here.
- Per-year counts cover only 2024 (2) and 2026 (4), with 2025 absent, so activity trends are two endpoints rather than a continuous series.