Early-career NLP researcher at HK PolyU: multimodal RAG benchmarks and text-to-SQL evaluation for LLMs
3
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
4
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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. Raw counts exist — 25 works across three years, 16 in 2026 — but nothing venue-normalized. This axis would measure publication volume, outlet quality, and output consistency; it stays unscored until that data is available.
Impact—
Not yet scored. OpenAlex reports h-index 4 and i10-index 3 — an early-career citation footprint too thin for a reliable measure. This axis would track citation counts, citing works, and field-normalized influence once the record matures.
Focus—
Not yet scored, though topic assignments exist: topic modeling (20), NLP techniques (12), database systems (6), IR and search behavior (4), multimodal ML (4). This axis would quantify topical concentration versus dispersion across the full publication record.
Collaboration56/100
Scored 56/100. The record links 16 distinct coauthors to 25 works over the last three years, pointing to team-based rather than solo research. Recurring names include Qinggang Zhang and Zheng Yuan (each appearing twice in the raw list), plus Xiao Huang, Feiran Huang, Jiayan Lin, Yujia Liu, Yilin Xiao, Hao Chen, Jiacheng Cai, and Huachi Zhou. All three catalog papers sit in middle authorship positions, consistent with specialist contributions inside larger groups.
Activity95/100
Scored 95/100 — the strongest axis here. countsByYear shows 5 works in 2024, 4 in 2025, then 16 in 2026, a roughly fourfold jump that lifts recent3yWorks to 25. All three Evolve-catalog papers are 2026 arXiv preprints, so the surge is already producing public artifacts. This pace places the researcher in the most active band of the underlying bibliometric distribution.
Reproducibility—
Not yet scored. Only one reviewed paper, CrossModalQA, carries a repro-readiness rating (medium); the two text-to-SQL papers have none. This axis would measure code, data, and benchmark release practices across the whole record.
Research summary
Zijin Hong is an early-career researcher at Hong Kong Polytechnic University working where natural language processing meets retrieval-augmented generation (RAG) and database-oriented language interfaces. The OpenAlex snapshot records 25 works over the last three years, accelerating sharply in 2026 (16 works versus 5 in 2024 and 4 in 2025), with an h-index of 4 and an i10-index of 3 — a footprint typical of the start of a publishing career. Topic assignments concentrate on topic modeling (20) and NLP techniques (12), followed by database systems and queries (6), information retrieval and search behavior (4), and multimodal machine learning (4). Three 2026 arXiv papers sit in the Evolve catalog, all in middle authorship positions: CrossModalQA, a cross-modal multi-hop benchmark for multimodal RAG (1,863 QA pairs, ~3.5 average hops, knowledge-graph-guided construction); an evaluation of LLMs on conversational text-to-SQL under chain ambiguity and intent drift; and a paradigm-level accuracy-cost analysis of in-context learning for text-to-SQL. The through-line is consistent: stress-testing how language models retrieve, ground, and execute knowledge across text, tables, and images. Collaboration is broad for the career stage — 16 distinct coauthors in three years, with Qinggang Zhang, Zheng Yuan, Xiao Huang, and Feiran Huang among recurring names.
Collaboration pattern
Collaboration is group-based: 16 distinct coauthors over three years against 25 works. The frequent-collaborator list is dominated by NLP and data-mining peers — Qinggang Zhang and Zheng Yuan each appear twice in the raw list, alongside Xiao Huang, Feiran Huang, Jiayan Lin, Yujia Liu, Yilin Xiao, Hao Chen, Jiacheng Cai, and Huachi Zhou. All three catalog papers place this researcher in middle authorship, none as corresponding author, consistent with specialist roles inside larger teams.
Research keywords
Highlights
CrossModalQA: a KG-guided cross-modal, multi-hop RAG benchmark
Stress-testing conversational text-to-SQL under chain ambiguity and intent drift
A 2026 arXiv study evaluating how LLMs handle multi-turn text-to-SQL when user intent drifts and ambiguity chains accumulate, adding an interface-robustness evaluation to the benchmark-oriented side of the catalog.
Paradigm-level accuracy-cost analysis of in-context learning for text-to-SQL
Asks whether in-context learning modules are worth their cost, comparing accuracy gains against inference expense at the paradigm level — an efficiency-oriented angle that runs through the text-to-SQL line of work.
Timeline
- 2024First year with recorded output in the OpenAlex snapshot: 5 works.
- 20254 works recorded — a comparatively quiet year before the 2026 acceleration.
- 202616 works recorded, the most productive year in the snapshot and roughly a fourfold jump over prior years; activity score reaches 95/100.
- 2026Three arXiv papers enter the Evolve catalog: CrossModalQA, conversational text-to-SQL under chain ambiguity and intent drift, and paradigm-level accuracy-cost analysis of in-context learning text-to-SQL — all middle authorship, none corresponding.
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.
- Bibliometrics (h-index, i10-index, counts by year, topics, coauthors) come from OpenAlex only; no other citation source was consulted.
- Only one of the three catalog papers (CrossModalQA) carries a full Evolve review, keywords, and a repro-readiness rating; the two text-to-SQL papers are listed without review data.
- The output, impact, focus, and reproducibility axes are unscored: early-career citation counts (h-index 4) and single-paper repro data are insufficient for reliable scoring.
- The frequent-coauthor list contains duplicate entries (Zheng Yuan and Qinggang Zhang each appear twice), so named-collaborator counts should be read with care; identity is name-resolved from OpenAlex records tied to ORCID 0009-0008-2283-255X.