Jia Wang

Xiamen University

Unclaimed · AI-structured profileOpenAlex

LLM Alignment Data Annotation and Human-AI Co-evolutionary Social Simulation Researcher at Xiamen University

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-08) 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, output rate, and first/corresponding authorship share over time. Only two arXiv (2026) cataloged papers are on file, with no publication-count or output-trend data, so no output score is computed yet.

Impact—

This axis would measure citation counts, h-index-style aggregates, and per-paper influence. No citation or impact data is available in the snapshot — both cataloged papers are recent 2026 arXiv preprints — so the impact axis is left unscored rather than estimated.

Focus0/100

The focus dimension scores 0/100, reflecting a perfectly even 50/50 split in the field distribution between LLM alignment/annotation NLP and LLM-based computational social science. Rather than one core line, the record shows two parallel tracks: onPanda addresses data annotation for alignment, while SocioVerse2 builds longitudinal social simulation. The zero score signals breadth across two adjacent subfields rather than an established single-field concentration.

Collaboration70/100

Evolve scores collaboration at 70/100, drawn from 33 distinct coauthors in the bibliometric record. The cataloged work is heavily team-based: both 2026 papers list the researcher in a middle author position with 12 recurring collaborators, including Xinnong Zhang, Xinyi Mou, Siyuan Wang, and Hanjia Lyu. This indicates an active, group-oriented style typical of large LLM-agent and social-simulation labs rather than solo or paired authorship.

Activity—

This axis would track publication recency and per-year activity cadence. The snapshot provides paper years only for two 2026 arXiv entries and no multi-year activity history, so cadence cannot be computed and the axis is left unscored.

Reproducibility55/100

Reproducibility scores 55/100, averaged over 2 cataloged papers (reproPapers = 2, reproMean = 55), both rated medium reproducibility readiness by Evolve. onPanda's annotation gains (a reported 52% time reduction versus post-editing) rest on small evidence with untested training benefit, and SocioVerse2's validation partially reuses the authors' companion studies. This yields a moderate score: methods are described but independent replication evidence remains thin.

Research summary

Jia Wang is a researcher at Xiamen University working at the intersection of natural language processing and computational social science, with a record that splits evenly (50/50) between LLM alignment and annotation methods and LLM-based social simulation. Evolve's catalog includes two 2026 arXiv papers with Wang in a middle author position. The first, onPanda, introduces token-level correction for annotating on-policy alignment data, reported to cut annotation time by 52% versus post-editing while keeping data on-policy; Evolve judges it a useful tool whose evidence base is small and whose training benefit is untested. The second, SocioVerse2, presents a longitudinal social-simulation framework under a human-AI co-evolutionary paradigm, evaluated across seven cases, though Evolve notes that validation partly repackages the authors' companion studies. Wang's collaboration footprint is broad: 33 distinct coauthors appear in the record, with a stable core of 12 frequent collaborators including Xinnong Zhang, Xinyi Mou, Siyuan Wang, and Hanjia Lyu, consistent with team-based LLM-agent research. Reproducibility averages 55/100 across the two rated papers, both at medium readiness. Citation counts, total output volume, and multi-year activity cadence are not yet available, so the impact, output, and activity axes remain unscored. This is an early-stage, name-resolved profile built from public bibliographic data.

Collaboration pattern

Wang works in large, rotating teams: the record shows 33 distinct coauthors, of whom 12 recur frequently — including Xinnong Zhang, Jiayu Lin, Xinyi Mou, Siyuan Wang, and Hanjia Lyu. Both cataloged 2026 papers sit in middle author positions with no corresponding-author role recorded, consistent with group projects at an LLM-agent and social-simulation lab. Evolve scores this collaboration pattern at 70/100.

Research keywords

LLM alignmenthuman-in-the-loop annotationtoken-level correctionon-policy dataSFT datapreference dataLLM-based social simulationagent-based modelingLLM agentshuman-AI collaborationlongitudinal simulationcounterfactual intervention

Highlights

Token-level correction for on-policy alignment data annotation
onPanda introduces token-level correction to annotate on-policy SFT and preference data for LLMs and agents, reported to cut annotation time by 52% versus post-editing. Evolve rates it a useful tool while noting the evidence base is small and training benefit untested; reproducibility readiness is medium.
Longitudinal social simulation under a human-AI co-evolutionary paradigm
SocioVerse2 proposes a longitudinal dynamic social simulation framework in which human and AI agents co-evolve, with counterfactual intervention support and a seven-case evaluation. Evolve calls the framework solid but notes validation partly repackages the authors' companion studies; reproducibility readiness is medium.
A dual track from alignment data tooling to computational social science
The cataloged record splits evenly between LLM alignment/annotation methods and LLM-based social simulation, with 33 distinct coauthors and a core of 12 frequent collaborators. Both 2026 papers hold middle author positions and medium reproducibility readiness, averaging 55/100.

Timeline

2026SocioVerse2, a longitudinal human-AI co-evolutionary social simulation framework, is released on arXiv and enters the Evolve catalog with medium reproducibility readiness.
2026onPanda, a token-level correction method for annotating on-policy alignment data, is released on arXiv; Evolve reports a 52% annotation-time reduction over post-editing.
2026Both cataloged papers carry middle author positions with no corresponding-author role, and the reproducibility average across the two papers stands at 55/100.

Papers in the Evolve catalog

Frequent co-authors

Xinnong Zhang×1 Jiayu Lin×1 Yixu Huang×1 Xinyi Mou×1 Yingqian Wu×1 Jingcong Liang×1 Shijun Lei×1 Jianing Shi×1 Guanying Li×1 Siyuan Wang×1 Hanjia Lyu×1 Zhenfei Yin×1

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.

Bibliometric scores and the coauthor network are computed from OpenAlex-derived public bibliometric data; no ORCID is linked, and 'Jia Wang' is a common name, so this profile is name-resolved from public bibliographic records and may merge or split identities.
Both cataloged papers are 2026 arXiv preprints; no peer-reviewed venue, citation counts, or publication-volume history are available, leaving output, impact, and activity axes unscored.
Reproducibility (55/100) is averaged over only two rated papers, so the estimate rests on a thin sample.
Field distribution covers only the two cataloged fields and may not reflect the researcher's full topical range.