Suncong Zheng — Tencent researcher linking NLP and text mining to LLM agent evaluation and benchmarking
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
12
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: this name-resolved record lacks a verified, de-duplicated publication list, so output cannot be computed. Once available, this axis would measure cumulative publication volume, annual productivity, and the share of first- or last-author papers.
Impact—
Not scored: the snapshot lists h-index 12 and i10-index 15, but field-normalized citation context is insufficient for a reliable impact measure. This axis would assess normalized citation rates, highly cited works, and venue-level influence.
Focus0/100
This profile shows near-total topical concentration. The OpenAlex topic profile is dominated by Topic Modeling (28 works) and Natural Language Processing Techniques (18), with Advanced Text Analysis (9), Text and Document Classification (5), and Sentiment Analysis (4) completing a single coherent text-mining cluster. The Evolve-reviewed 2026 papers sit in one field, machine learning (LLM agent evaluation), so the current line extends this NLP base into agent benchmarking with no sign of scattering.
Collaboration89/100
The record lists 25 distinct coauthors, a moderately broad network consistent with the collaboration score of 65. Twelve recurring collaborators anchor it, including Ming Zhang, Zhenghao Xiang, Peizhong Gao, Yujiong Shen, Shihan Dou, Zhangyue Yin and Junjie Ye — a roster suggesting a stable, lab-centered cluster. Both reviewed 2026 papers carry middle authorship, fitting a contributor role inside multi-author industry teams rather than solo work.
Activity86/100
Activity is high and accelerating: the snapshot records 18 works over the last three years, with 7 in 2024, 3 in 2025 and 8 already in 2026 — matching the 8-work peak of 2016. After single-work years in 2020, 2022 and 2023 and a gap across 2018-2019, the 2024-2026 run marks a clearly busier second phase of research output.
Reproducibility43/100
Reproducibility scores 55, from a thin base (reproMean 55 across 1 scored paper). Evolve's verdicts temper enthusiasm: ExplorationBench is rated low readiness (n=3 trajectories, single-run controls), and DualViewEval medium (strong five-benchmark results but tiny agent pools, no significance testing). Design rigor is evident — exact grading, dual outcome/process views — yet small samples are the recurring weakness, and artifact availability is not confirmed in the data.
Research summary
Suncong Zheng is a researcher at Tencent (China) whose public record, name-resolved from OpenAlex, spans natural language processing and text mining with a recent, active turn toward LLM agent evaluation. The topical profile is heavily concentrated: Topic Modeling (28 works) and Natural Language Processing Techniques (18) dominate, followed by Advanced Text Analysis (9), Text and Document Classification (5), and Sentiment Analysis (4). The snapshot lists an h-index of 12 and an i10-index of 15. Activity has re-accelerated sharply: 18 works in the last three years, with 7 in 2024, 3 in 2025 and 8 in 2026, matching the record's 2016 peak. Evolve has reviewed two 2026 arXiv papers, both with middle authorship: ExplorationBench, a verifiable alien-world benchmark for AI exploration with exact grading that is limited by small samples (n=3 trajectories), and DualViewEval, which applies dual-view relational learning to compress agent benchmarks across five testbeds. Evolve's reproducibility assessment is cautious — one scored paper averaging 55, with readiness ratings of low and medium — citing small agent pools and missing significance tests. The overall picture is a focused NLP specialist now contributing to emerging agent-evaluation research within collaborative teams.
Collaboration pattern
Suncong Zheng works in team-based industry research: 25 distinct coauthors appear in the record, with 12 frequent collaborators — including Ming Zhang, Zhenghao Xiang, Peizhong Gao, Yujiong Shen, Shihan Dou, Zhangyue Yin and Junjie Ye — suggesting a stable lab-centered cluster. Both Evolve-reviewed 2026 papers are middle-author contributions, indicating a contributor role within larger groups rather than solo or corresponding-author leadership.
Research keywords
Highlights
ExplorationBench: probing AI exploration in verifiable alien worlds
Dual-view relational learning for cheaper agent benchmarking
A text-mining core now pointed at LLM agents
The OpenAlex topic profile concentrates on Topic Modeling (28 works), Natural Language Processing Techniques (18), Advanced Text Analysis (9), Text and Document Classification (5) and Sentiment Analysis (4). The two 2026 reviewed papers extend this NLP base into LLM agent evaluation, marking a deliberate pivot within one skill set.
Timeline
- 2013Earliest indexed work in the OpenAlex record (1 publication); the record stays sparse through 2015 (2 works).
- 2016Early productivity peak: 8 indexed works, the joint-highest annual output in the record.
- 20175 indexed works close the first active phase; the snapshot shows no indexed output in 2018-2019.
- 2024Renewed activity begins: 7 indexed works, after single-digit years from 2020 to 2023.
- 2026Current peak: 8 indexed works so far, including both Evolve-reviewed papers on LLM agent evaluation (ExplorationBench and Dual-View Relational Learning).
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.
- Bibliometric indicators are drawn from OpenAlex only; no ORCID is linked, so the record is name-resolved and may merge or split identically named authors.
- Output and impact dimension scores are not computed; h-index 12 and i10-index 15 are raw snapshot values without field-normalized citation context.
- Only two Evolve-reviewed papers exist, both 2026 arXiv preprints with middle authorship; earlier work is visible only through aggregated topic and year counts.
- Annual counts have gaps (no indexed works in 2014 and 2018-2019), so long-term productivity trends should be read cautiously.