Yongliang Shen

Zhejiang University of Science and Technology

Unclaimed · AI-structured profileOpenAlex

NLP researcher applying LLM post-training and data-efficient instruction tuning to telecom reasoning and Text-to-SQL.

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—

Not yet scored: no publication-count or productivity metrics are available in this snapshot. This axis would measure cumulative and per-year paper output across venues, which cannot be computed from the two reviewed papers alone.

Impact—

Not yet scored: citation counts, h-index, and downstream-usage signals are absent from this snapshot. This axis would measure how widely the work is cited and adopted; neither reviewed paper has citation data available yet.

Focus0/100

The focus axis scores 0 out of 100. The field distribution splits evenly — 50% natural language processing (Text-to-SQL / semantic parsing) and 50% natural language processing / large language models — indicating two parallel research threads rather than one deep specialization. With only two catalog-reviewed papers, both from 2026, the sample is too small for a stable focus signature, so this score is best read as early-stage breadth.

Collaboration54/100

The collaboration axis scores 54 out of 100, based on 14 distinct co-authors observed in the bibliometric snapshot. Eleven recurring names appear — including Haoyuan Ma, Weiming Lu, Bohao Wang, Hang Zou, Lina Bariah, Li Wei and Chongwen Huang — spanning both the telecom and Text-to-SQL threads. A mid-range score indicates a genuine but still moderately sized network, and both catalog papers place the researcher in a middle, non-corresponding author position.

Activity—

Not yet scored: recent publication cadence and year-by-year volume cannot be derived from this snapshot. This axis would measure ongoing research tempo; only the 2026 arXiv submissions are visible, so no trend over time is established.

Reproducibility55/100

The reproducibility axis scores 55 out of 100, a mean computed over 2 reviewed papers, each rated medium readiness. TelecomGPT-R1 reports solid ablations but carries contamination and evaluation-setting caveats; LIMIT is well ablated yet faces schema-overlap and statistical-significance concerns. The mid-range score reflects reproducible pipelines whose evaluation conditions still warrant independent verification before headline results are treated as settled.

Research summary

Yongliang Shen is a natural language processing researcher affiliated with Zhejiang University of Science and Technology whose reviewed work concentrates on two threads: large language model post-training for domain reasoning, and data-efficient instruction tuning for Text-to-SQL. The Evolve catalog holds two 2026 arXiv papers, both with middle-author, non-corresponding positions. TelecomGPT-R1 applies reinforcement learning with verifiable rewards (DAPO) and chain-of-thought distillation to unify reasoning across heterogeneous telecom tasks, reporting 89.6% mean on GSMA Open Telco with claims of beating GPT-5 and Claude — solid ablations, though contamination and evaluation-setting caveats persist. LIMIT shows that 796 curated samples give Qwen3-8B 69.1% BIRD execution accuracy, using genetic-algorithm data selection and LLM-as-judge filtering, with schema-overlap and significance concerns flagged. The field distribution splits evenly (50% Text-to-SQL / semantic parsing, 50% NLP / LLM), and the radar focus score of 0 reflects this two-thread breadth rather than deep specialization. Collaboration scores 54 with 14 distinct co-authors recorded. Reproducibility averages 55 across 2 papers, both rated medium readiness. Output, impact, and activity axes remain unscored for lack of underlying data. This is an early-stage, name-resolved profile built from a bibliometric snapshot plus Evolve's reviews; the researcher's own name also appears in the frequent-coauthor list, indicating possible disambiguation noise.

Collaboration pattern

The collaboration axis scores 54/100 with 14 distinct co-authors in the snapshot. Eleven recurring names appear — Haoyuan Ma, Hengwei Liu, Linjuan Wu, Weiming Lu, Bohao Wang, Chenwei Wu, Hang Zou, Yu Tian, Lina Bariah, Li Wei and Chongwen Huang — spanning both the telecom and Text-to-SQL threads. Both catalog papers list the researcher as a middle, non-corresponding author, so current evidence shows team participation rather than project leadership.

Research keywords

LLM post-trainingreinforcement learning with verifiable rewardsDAPOdomain-specific LLMtelecommunicationschain-of-thought distillationText-to-SQLinstruction tuningdata selectiondata efficiencyLLM-as-judgegenetic algorithm

Highlights

TelecomGPT-R1: one post-training pipeline for heterogeneous telecom reasoning
Unified post-training combining reinforcement learning with verifiable rewards (DAPO) and chain-of-thought distillation yields a reported 89.6% mean on GSMA Open Telco, claimed to beat GPT-5 and Claude. Evolve rates reproducibility medium: ablations are solid, but contamination and evaluation-setting caveats persist.
LIMIT: less is more for Text-to-SQL instruction tuning
A data-selection pipeline combining a genetic algorithm with LLM-as-judge filtering curates 796 samples that give Qwen3-8B 69.1% execution accuracy on BIRD. Evolve rates reproducibility medium: the pipeline is well ablated, but schema-overlap and statistical-significance concerns remain.

Timeline

2026TelecomGPT-R1 appears on arXiv: unified post-training combining reinforcement learning with verifiable rewards (DAPO) and chain-of-thought distillation for reasoning across heterogeneous telecom tasks.
2026LIMIT appears on arXiv, showing 796 curated instruction-tuning samples drive Qwen3-8B to 69.1% execution accuracy on BIRD for Text-to-SQL.
2026Both papers enter the Evolve catalog with middle-author positions and medium repro-readiness ratings, establishing two parallel research threads within the same year.

Papers in the Evolve catalog

Frequent co-authors

Haoyuan Ma×1 Hengwei Liu×1 Linjuan Wu×1 Weiming Lu×1 Yongliang Shen×1 Bohao Wang×1 Chenwei Wu×1 Hang Zou×1 Yu Tian×1 Lina Bariah×1 Li Wei×1 Chongwen Huang×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.

Radar scores and co-author counts are computed from a public bibliometric snapshot; citation, venue and productivity metrics are not part of the supplied data.
Both reviewed papers are 2026 arXiv preprints and may not yet be peer reviewed.
Output, impact and activity axes are unscored because the underlying bibliometric data is insufficient.
'Yongliang Shen' is a common name and appears even within this profile's own frequent-coauthor list; the profile is name-resolved from public bibliographic data and may contain disambiguation noise.