Sergio A. Diaz

National Yang Ming Chiao Tung University · Taiwan

Unclaimed · AI-structured profileOpenAlex

KGQA researcher at NYCU (Taiwan): traceable multi-hop graph navigation and evaluation beyond answer accuracy

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-15) 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: reliable publication-volume data is unavailable in this snapshot. This axis would measure total research output across venues and years — including preprints, peer-reviewed papers, datasets, and software — to gauge sustained production.

Impact—

Not yet scored: citation-based evidence is insufficient — h-index and i10-index are recorded as 0 with no citation time series. This axis would measure citations, uptake of released benchmarks, and influence on follow-up work.

Focus0/100

The focus score is 0: recorded output splits evenly (0.5 and 0.5) across two adjacent NLP subfields — knowledge graphs and KGQA/LLM reasoning — rather than concentrating in one. Topic signals reinforce the breadth: topic modeling, semantic web and ontologies, and advanced graph neural networks each appear twice. In practice both cataloged papers address the same KGQA navigation question, so the low score reflects balanced dual labeling more than scattered interests.

Collaboration32/100

The collaboration score is 32, based on four distinct coauthors in the snapshot. Both cataloged papers list him as a middle author without corresponding-author status, consistent with contributing within a small team. Recurring coauthors are Eduin E. Hernandez, Luis F. Garcia, Nurassyl Askar, and Stefano Rini, pointing to a compact, likely single-group circle at National Yang Ming Chiao Tung University rather than a broad international network.

Activity32/100

Activity scores 32. The counts-by-year record shows exactly two works in 2026 and no earlier years in the window, with recent3yWorks at 2. Both items are arXiv preprints released in the same year, indicating a concentrated burst of output rather than a sustained multi-year cadence — consistent with an early-stage record whose long-term publishing rhythm is not yet established.

Reproducibility78/100

Reproducibility is the strongest axis at 78, averaged over the two reviewed papers. The THESEUS paper rates high for readiness — question-conditioned navigation with gold paths and fidelity metrics — while the SLM evaluation paper rates medium, with small benchmarks and no significance tests noted. A high mean on a two-paper base signals sound methodological reporting, though the evidence base remains small.

Research summary

Sergio A. Diaz is a researcher at National Yang Ming Chiao Tung University in Taiwan whose OpenAlex-recorded work sits in natural language processing over knowledge graphs, split evenly (0.5 and 0.5) between knowledge-graph research and KGQA with LLM reasoning. The Evolve catalog holds two 2026 arXiv preprints on which he is a middle author. 'Theseus in the Graph' frames knowledge graph question answering as question-conditioned graph navigation with gold paths and fidelity metrics; Evolve judged the benchmark solid while noting the closest prior baselines were excluded, and rated reproducibility readiness high. 'The Path Matters' evaluates small language models beyond answer accuracy, showing that path fidelity and answer accuracy can rank models differently; Evolve called the design sound but flagged small benchmarks and missing significance tests, rating readiness medium. The two papers average a reproducibility score of 78. Early-stage indicators: h-index and i10-index are both 0, two works fall in the recent three-year window, all dated 2026, with four distinct coauthors — Eduin E. Hernandez, Luis F. Garcia, Nurassyl Askar, and Stefano Rini. Topic modeling, semantic web and ontologies, and advanced graph neural networks each appear twice among topic signals.

Collaboration pattern

Sergio A. Diaz's recorded collaboration circle is compact: four distinct coauthors appear in the snapshot, with Eduin E. Hernandez, Luis F. Garcia, Nurassyl Askar, and Stefano Rini recurring. On both cataloged 2026 papers he is a middle author without corresponding status, consistent with contributing within a small, stable team — plausibly based at National Yang Ming Chiao Tung University — rather than leading multi-group consortia.

Research keywords

knowledge graph question answeringmulti-hop reasoninggraph navigationpath fidelityPath Edit Distancesmall language modelstraceabilityevaluation methodologyreinforcement learning agentssemantic webtopic modelinggraph neural networks

Highlights

THESEUS: KGQA as traceable graph navigation
Frames knowledge graph question answering as question-conditioned navigation over the graph, introducing gold paths and path-fidelity metrics for traceable multi-hop reasoning. Evolve judged the benchmark solid while noting the closest prior baselines were excluded, and rated reproducibility readiness high.
Evaluating small language models beyond answer accuracy
A controlled study of small language models on KGQA showing that path fidelity and answer accuracy can rank models differently. Evolve found the design sound but flagged small benchmarks and no significance tests; reproducibility readiness is rated medium.
A consistent path-fidelity methodology across both papers
Both 2026 papers share one thread — scoring a model's navigation path against a gold path (e.g., via Path Edit Distance) rather than only the final answer — and together carry a mean Evolve reproducibility score of 78, with readiness rated high and medium respectively.

Timeline

2026Published 'Theseus in the Graph: Towards Traceable Multi-Hop Graph Navigation' on arXiv, framing KGQA as question-conditioned graph navigation with gold paths.
2026Published 'The Path Matters: Evaluating Small Language Models Beyond Answer Accuracy in KGQA' on arXiv, a controlled study of SLM navigation fidelity versus answer accuracy.
2026Both papers entered the Evolve review catalog with reproducibility-readiness ratings of high and medium, contributing to a mean reproducibility score of 78.

Papers in the Evolve catalog

Frequent co-authors

Eduin E. Hernandez×2 Luis F. Garcia×2 Nurassyl Askar×2 Stefano Rini×2

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 come from an OpenAlex-based snapshot; h-index and i10-index are recorded as 0, and the output and impact axes lack sufficient data to score.
The Evolve catalog covers only two 2026 arXiv preprints; no peer-reviewed venues, citation counts, or funding information are captured.
No ORCID is linked, so this profile is name-resolved from public bibliographic data and could conflate identically named authors.
The reproducibility score of 78 reflects Evolve's assessment of the two reviewed papers only, not a general measure of all work.