Tencent researcher in graph-based retrieval-augmented generation and diffusion language models for multi-hop reasoning
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
10
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-13) 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—
Output volume is not yet scored: this snapshot lacks the underlying totals — cumulative publication count and its distribution across venues and years of record. This axis would measure total works produced and the long-run productivity trajectory of the researcher.
Impact—
Impact is not yet scored: citation counts, citation velocity, and venue-level influence are missing from the current data. This axis would measure how widely the work is cited and the influence it exerts on the field.
Focus0/100
The focus axis scores 0/100, reflecting a maximally split field distribution: 50% machine learning / natural language processing and 50% machine learning (graph representation learning and retrieval-augmented generation). The five counted topics — topic modeling (15), an off-profile legal cluster (9), NLP techniques (8), domain adaptation and few-shot learning (5), and text classification (5) — span several adjacent NLP subareas plus one outlier cluster, consistent with low measured concentration.
Collaboration59/100
Collaboration scores 59/100, computed from 18 distinct coauthors. The recurring-partner list centers on Senlei Zhang, Junnan Dong, Xing Sun, Linhao Luo, Siyu An, Shuhao Zhang, and Gong Chen (several names repeat in the source list, and the list also contains the researcher's own name — an indexing artifact). Both catalog papers are middle-author team efforts rather than solo or corresponding-authored work, indicating a collaborative contributor pattern.
Activity64/100
Activity scores 64/100. The per-year record shows 2 works in 2014 rising through 4 in 2016 to a peak of 9 in each of 2020 and 2021, then 5 in 2022, 5 in 2023, a dip to 2 in 2024, a rebound to 4 in 2025, and 2 recorded in 2026. With 8 works in the last three years, the pace remains sustained and currently active.
Reproducibility38/100
Reproducibility scores 38/100, a mean taken across the 2 cataloged papers (reproPapers 2, reproMean 38). LADDER is rated medium readiness and WFM low. Both share the same flagged gap: headline numbers — +5.6 EM with a 4.1x speedup for LADDER, a claimed 10.5x speedup with SOTA QA gains for WFM — lack significance tests, and WFM additionally carries an unexplained 'planning values' note.
Research summary
Qian-Wen Zhang is a researcher at Tencent (China) whose OpenAlex-resolved record sits in machine learning for natural language processing, split evenly (0.5/0.5) between general ML/NLP and graph representation learning with retrieval-augmented generation. Reported indicators include an h-index of 10, an i10-index of 10, and 8 works in the last three years. Annual counts peak at 9 works in each of 2020 and 2021, then settle to 5 (2022), 5 (2023), 2 (2024), 4 (2025), and 2 so far in 2026. Two 2026 arXiv preprints anchor the Evolve catalog, both middle-author and non-corresponding: LADDER, a graph-guided diffusion language model reporting +5.6 exact match and a 4.1x speedup for multi-hop QA, and WFM, a wiki foundation model over LLM-Wiki graphs claiming a 10.5x speedup and SOTA QA gains. Evolve's reproducibility assessment averages 38 across these 2 papers (LADDER medium, WFM low), with missing significance tests the recurring weakness, plus an unexplained 'planning values' note in WFM. An 18-distinct-coauthor footprint points to steady team research. The topic list also contains an off-profile legal cluster, flagged as possible name-ambiguity noise.
Collaboration pattern
With 18 distinct coauthors, Zhang works in team settings. The frequent-coauthor list centers on Senlei Zhang, Junnan Dong, Xing Sun, Linhao Luo, Siyu An, Shuhao Zhang, and Gong Chen. Both 2026 catalog papers are middle-author, non-corresponding contributions, suggesting a specialist contributor within larger groups on these outputs. The source list repeats several names and includes Zhang's own name, an indexing artifact.
Research keywords
Highlights
LADDER: graph-guided diffusion language models for multi-hop reasoning (+5.6 EM, 4.1x speedup claimed)
WFM: a wiki foundation model over LLM-Wiki graphs for complex agentic reasoning
Timeline
- 2014Earliest works in the OpenAlex record: 2 publications in 2014, followed by 1 in 2015.
- 2016Output rises to 4 works in 2016 in the early phase of the record.
- 2020Peak productivity: 9 works in 2020, matched with another 9 in 2021.
- 2022Sustained mid-pace output of 5 works in 2022 and 5 in 2023, before a dip to 2 in 2024.
- 2026Two arXiv preprints enter the Evolve catalog — LADDER and WFM, both on graph-based retrieval-augmented generation and multi-hop reasoning, as middle-author contributions.
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.
- All bibliometrics come from a single OpenAlex snapshot (h-index 10, i10-index 10, per-year counts); no other citation source is included.
- The topic distribution includes 'International Arbitration and Investment Law' (9 works), which is off-profile for an ML/NLP researcher and likely reflects name-ambiguity noise; this profile is name-resolved from public bibliographic data.
- Output and impact dimensions are unscored because the underlying data (total publication counts, citations) is missing from the snapshot.
- Both cataloged papers are 2026 arXiv preprints with middle-author positions; no peer-reviewed venue status or author-contribution detail is available.