Philip S. Yu

University of Illinois Chicago

Unclaimed · AI-structured profileOpenAlex

LLM evaluation and self-evolving agents researcher at UIC; paired-design rigor in legal NLP and time series analysis

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-02) 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 scored yet. The snapshot lacks cumulative publication counts: h-index and i10-index are recorded as zero and yearly coverage spans only 2026 (2 works). This axis would measure total publication volume and career-long productivity trend.

Impact—

Not scored yet. No citation-level metrics — citations, h-index, influential citations — are available for this OpenAlex identity in the snapshot. This axis would measure citation impact, uptake in follow-up work, and scholarly visibility over time.

Focus0/100

The radar scores focus at 0/100, reflecting an unusually even portfolio: the field distribution splits 50/50 between machine learning (LLM agents for time series analysis) and natural language processing (legal NLP, model evaluation). OpenAlex topics scatter across five areas — topic modeling (2), language and cultural evolution, time series forecasting, authorship attribution, and AI in law — so current evidence shows two parallel research lines rather than one dominant specialty.

Collaboration65/100

Collaboration scores 65/100. The recent window records 25 distinct co-authors against only 2 works, an unusually high co-author density that signals lab-scale group work. The frequent-coauthor list spans ten recurring names — including Jie Yang, Yan Zheng, Jiarui Sun, Xiran Fan, and Junpeng Wang — and the two reviewed papers place the researcher in the middle and last author positions, consistent with a senior supervisory role within rotating project teams.

Activity32/100

Activity scores 32/100. The OpenAlex-derived snapshot shows a thin recent record: recent3yWorks is 2, and countsByYear contains only 2026 with 2 works, both arXiv preprints. Both papers appeared in a single year, so the low score largely reflects incomplete source coverage rather than confirmed inactivity — career-long output is simply not visible in this window.

Reproducibility55/100

Reproducibility scores 55/100, computed as a mean of 55 across 2 reviewed papers. Evolve rates both as medium reproducibility readiness: the legal-NLP study runs a rigorous paired design on ContractNLI but rests on a 30-anchor panel that keeps stability claims exploratory, while TimeEvo's paired admission gate covers all 30 task-backbone cells yet omits self-evolving baselines.

Research summary

Philip S. Yu is a researcher at the University of Illinois Chicago whose profile is name-resolved from OpenAlex bibliographic records — a necessary caveat given how common the name is. The current snapshot centers on two 2026 arXiv preprints reviewed by Evolve, spanning a deliberately split portfolio: machine learning work on LLM agents for time series analysis (50% of the field distribution) and natural language processing work on legal NLP and model evaluation (50%). On the first line, the last-authored TimeEvo paper proposes failure-driven self-evolution of a time series agent, synthesizing a tool library admitted through a paired gate and reportedly winning all 30 task-backbone cells, though no self-evolving baselines were run. On the second, the middle-authored 'Same Scores, Different Decisions' study evaluates JEV against nine language models on ContractNLI, exposing cost-latency-correctness trade-offs where identical scores lead to different decisions; the paired design is rigorous, but a 30-anchor panel keeps stability claims exploratory. OpenAlex topics scatter across five areas, including topic modeling (2) and AI in law (1). Collaboration stands out: 25 distinct co-authors against only 2 recent works. Evolve rates both papers at medium reproducibility readiness (reproMean 55), while output and impact remain unscored because the snapshot's h-index and i10-index are unpopulated.

Collaboration pattern

Collaboration is the strongest measurable axis (65/100): 25 distinct co-authors appear against only 2 recent works, pointing to lab-scale teamwork. Ten recurring co-authors include Jie Yang, Yan Zheng, Jiarui Sun, Xiran Fan, Junpeng Wang, Liang Wang, Zelin Xu, Qinghua Liu, Zhengyu Fang, and Yiwei Cai. The reviewed papers place the researcher in a middle position on the legal-NLP study and last on TimeEvo, consistent with a senior supervisory role; roles are inferred from author positions, not confirmed.

Research keywords

LLM agentself-evolving agenttool library synthesistime series analysisanomaly detectionlegal document understandingnatural language inferenceLLM evaluationcost-accuracy trade-offprompt sensitivity

Highlights

Same scores, different decisions: JEV vs. nine LLMs on ContractNLI
This middle-author arXiv study evaluates JEV and nine language models on legal document understanding over ContractNLI, exposing cost-latency-correctness trade-offs where identical scores lead to different decisions. Evolve calls the paired design rigorous but notes the 30-anchor panel keeps stability claims exploratory; reproducibility readiness is rated medium.
TimeEvo: failure-driven self-evolution of a time series agent
The last-authored TimeEvo preprint introduces failure-driven self-evolution for an LLM time series agent: synthesized tool libraries are admitted through a paired admission gate, reportedly winning all 30 task-backbone cells. Evolve rates the protocol rigorous but flags that no self-evolving baselines were compared; readiness is medium.
A common signature: paired designs with explicit admission criteria
Across both 2026 Evolve-reviewed papers, comparisons are paired and claims are scoped to the tested panels — a consistency reflected by the 55/100 reproducibility score and the two medium readiness ratings, even as anchor-panel and baseline gaps temper the conclusions.

Timeline

2026Posted 'Same Scores, Different Decisions' (arXiv, middle author): a paired evaluation of JEV and nine LLMs on ContractNLI for legal document understanding.
2026Posted 'TimeEvo: Failure-Driven Self-Evolution of a Time Series Agent' (arXiv, last author): tool library synthesis admitted via a paired gate, winning all 30 task-backbone cells.
2026Both papers entered the Evolve catalog at medium reproducibility readiness; the snapshot records 2 works in 2026, 25 distinct co-authors, and a reproMean of 55.

Papers in the Evolve catalog

Frequent co-authors

Jie Yang×1 Yan Zheng×1 Jiarui Sun×1 Xiran Fan×1 Junpeng Wang×1 Liang Wang×1 Zelin Xu×1 Qinghua Liu×1 Zhengyu Fang×1 Yiwei Cai×1 Philip S. Yu×1 Zhengyu Fang×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 are derived from OpenAlex only; 'Philip S. Yu' is a common name, so this profile is name-resolved from public bibliographic records and may merge or omit works.
h-index and i10-index are recorded as 0 in the snapshot (likely unpopulated), so the output and impact axes are not scored.
Coverage is thin: countsByYear contains only 2026 (2 works) and both catalog papers are 2026 arXiv preprints; career-long history is not represented.
The frequent-coauthor list includes the researcher's own name and one duplicated entry; co-author roles are inferred from author positions only.