Zhen Wang

University of California San Diego · United States

Unclaimed · AI-structured profileOpenAlex

UC San Diego last-author on LLM work spanning human-aware dialogue and multi-agent scientific hypothesis discovery

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-20) 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—

Not yet scored: the OpenAlex snapshot shows no usable productivity signal for this record — citation indices read zero and coverage is limited to 2026. This axis would measure publication volume and per-year output across venues.

Impact—

Not yet scored: the two 2026 arXiv preprints have no citation record in the snapshot, so influence cannot be measured. This axis would track citations, citing works, and uptake of these methods by other research groups.

Focus0/100

The focus score is 0, the weakest of the four computed axes, and the field distribution explains why: work is split evenly, 50% natural language processing and 50% AI-for-science LLM agents. Topic tags reinforce a deliberately broad portfolio — artificial intelligence in healthcare and education and explainable AI at 4 counts each, multimodal ML and materials-science ML at 2 each — rather than one narrow specialty.

Collaboration69/100

Collaboration is the strongest computed axis at 69/100. The record lists 31 distinct co-authors over the recent three-year window against only 4 works — a wide author network per paper. Frequent co-authors include Zhiting Hu, Trey Ideker, Wei Wang, Dexter Pratt, and Pratibha Jagannatha, spanning the NLP and biomedical sides of the portfolio. Both cataloged papers are multi-author efforts with this researcher in the last position.

Activity47/100

Activity scores a moderate 47/100. All recorded output is recent: countsByYear contains a single year, 2026, with 4 works, and recent3yWorks is also 4. There is no prior-year publication record in the snapshot, so the score reflects a concentrated burst of preprints rather than a long sustained cadence. Both cataloged papers appeared on arXiv in 2026.

Reproducibility20/100

Reproducibility is the lowest scored axis at 20/100, based on 2 evaluated papers (reproMean 20). Evolve rates both Mind2Dialogue and HypoEvolve 'low' readiness: the former distills from a closed Oracle teacher, and the latter leaves baseline gaps and a control-condition caveat unresolved. Independent replication would require released code, data, and evaluation pipelines for both systems.

Research summary

Zhen Wang is a researcher at the University of California San Diego whose profile is name-resolved from public bibliographic data (OpenAlex). The cataloged record is small but recent: two 2026 arXiv preprints on which Wang is the last author, spanning a balanced split of natural language processing (50%) and AI for science with LLM agents (50%). Mind2Dialogue trains human-aware language models by simulating user mental states through an Oracle teacher, and HypoEvolve uses a generational genetic algorithm over multi-agent LLM systems to surface drug-repurposing hypotheses. Topic tags reinforce this breadth: artificial intelligence in healthcare and education and explainable AI (4 mentions each), plus multimodal machine learning and machine learning in materials science (2 each). Collaboration is the strongest computed radar axis (69/100), with 31 distinct co-authors over the recent three-year window against only 4 recorded works; frequent co-authors include Zhiting Hu, Trey Ideker, Wei Wang, and Dexter Pratt. Activity is moderate (47), focus is split (0), and reproducibility is low (20, both papers rated low readiness). Citation indices read zero in this snapshot, so output and impact are flagged missing rather than scored.

Collaboration pattern

The record shows a wide, thin collaboration pattern: 31 distinct co-authors across only 4 recent works. Named frequent co-authors include Zhiting Hu, Jieyuan Liu, Mengzhou Hu, Jefferson Chen, JungHo Kong, Pratibha Jagannatha, Yiming Gao, Dexter Pratt, Hsin-Yuan Lee, Trey Ideker, and Wei Wang. On both cataloged papers Wang takes the last author position, consistent with a senior or supervisory role, though neither lists Wang as corresponding author.

Research keywords

human-aware language modelstheory of minduser simulationpersonalizationsynthetic dialogue dataprivileged information distillationscientific hypothesis discoverymulti-agent LLM systemsgenetic algorithmdrug repurposingevolutionary searchexplainable AI (XAI)

Highlights

Genetic algorithms coordinate LLM agents for drug-repurposing hypothesis discovery
A generational genetic algorithm coordinating multiple LLM agents lifted drug-repurposing hypothesis scores above six baselines. Evolve's review flags unresolved issues: the margin over the strongest baseline is unsettled on one metric, and a constant-drug control won on held-out Open Targets data.
Training human-aware language models from a simulated-mind Oracle
Mind2Dialogue distills an Oracle teacher that sees simulated user mental states into assistants, with personalization gains holding across three backbones. Evolve notes the approach rests on a closed teacher model and rates its reproducibility readiness low.
Last-author seniority across a two-field 2026 portfolio
Both 2026 cataloged preprints list Wang as last author — one in natural language processing, one in AI for science — mirroring an even 50/50 field split and a wide 31-co-author network, though neither paper names Wang as corresponding author.

Timeline

2026HypoEvolve, a genetic-algorithm framework over multi-agent LLMs for scientific hypothesis discovery in drug repurposing, is posted to arXiv with Wang as last author.
2026Mind2Dialogue, on training human-aware language models by simulating user mental states, is posted to arXiv, also with Wang as last author.
2026OpenAlex counts 4 works for this record in 2026 — its only year with recorded output — alongside 31 distinct co-authors in the recent three-year window.
2026Evolve reviews both preprints, rating each 'low' on reproducibility readiness and noting open questions around closed teachers, baseline gaps, and control conditions.

Papers in the Evolve catalog

Frequent co-authors

Zhiting Hu×2 Zhen Wang×2 Jieyuan Liu×1 Mengzhou Hu×1 Jefferson Chen×1 JungHo Kong×1 Pratibha Jagannatha×1 Yiming Gao×1 Dexter Pratt×1 Hsin-Yuan Lee×1 Trey Ideker×1 Wei Wang×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.

Bibliometrics come from OpenAlex only; 'Zhen Wang' is a common name and the record is name-resolved from public bibliographic data, so attribution may include noise — a frequent-coauthor entry even duplicates the researcher's own name.
The snapshot reports h-index 0 and i10-index 0 with coverage in 2026 only; output and impact axes are flagged as missing rather than scored.
Both cataloged papers are 2026 arXiv preprints with no peer-reviewed venue shown in the data.
Reproducibility scoring covers only 2 evaluated papers, both rated low readiness, so the 20/100 figure rests on a small base.