Siyu An — LLM agent long-term memory and graph foundation models; early-stage profile from two reviewed 2026 arXiv preprints
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-20) 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: the current bibliometric snapshot lacks sufficient publication-volume data for this researcher. This axis would measure total output — paper count, venue quality, and first/corresponding-author share — beyond the two 2026 arXiv preprints reviewed here.
Impact—
Not yet scored: citation and uptake data is unavailable in this snapshot. This axis would measure citations, downstream adoption, and community influence of the reviewed 2026 preprints on agent memory and graph foundation models.
Focus0/100
Scores 0 on focus, and the raw field distribution explains why: an exact 50/50 split between natural language processing (LLM agents, long-term memory and personalization) and machine learning (graph representation learning and retrieval-augmented generation). Both reviewed 2026 papers straddle this boundary, pairing graph neural networks with RAG. Read as deliberate interdisciplinary positioning around graph-augmented agents rather than a drifting agenda, though no single field yet dominates the record.
Collaboration56/100
Moderate score of 56, built on 16 distinct coauthors across the reviewed work, with 12 named frequent collaborators — including Junnan Dong, Ruizhi Qiao, Xing Sun, and Ruifeng Xu — suggesting a stable core team. On both 2026 preprints Siyu An sits in a middle author position without the corresponding-author role, consistent with a contributing specialist within larger groups rather than solo or lead-author research.
Activity—
Not yet scored: year-by-year publication records are too sparse to compute a cadence. This axis would measure publishing consistency and recent activity over time; only the two 2026 arXiv preprints are currently anchored in the data.
Reproducibility20/100
Score of 20 — the mean over 2 reviewed papers, both rated low reproducibility readiness. The reviews are specific: the latent neuro-symbolic memory paper lacks significance tests, a stated judge, and code, while WFM shows absent statistics and a leaked 'planning values' note. The methods look promising, but releasing code, evaluation protocols, and statistical rigor would directly lift this axis.
Research summary
Siyu An is an early-stage researcher whose reviewed work sits at the interface of natural language processing and machine learning, split evenly (50/50 in the field distribution) between LLM agents — long-term memory and personalization — and graph representation learning with retrieval-augmented generation. The Evolve catalog contains two 2026 arXiv preprints. The first, Disentangling Long-Term Memory via Latent Neuro-Symbolic Reasoning, proposes query-conditioned latent graphs for personalized agents and reports large personalization-benchmark gains. The second, WFM: Wiki Foundation Model for Complex Agentic Reasoning, builds a graph foundation model over dense LLM-Wiki graphs with a claimed 10.5x speedup and state-of-the-art multi-hop QA results. Both are reviewed as promising but under-verified: missing significance tests, an unstated judge, no code release, and — for WFM — a leaked 'planning values' note and absent statistics, yielding a low reproducibility score of 20 (the mean over 2 reviewed papers). Collaboration is moderate (score 56), with 16 distinct coauthors and a recurring core of 12 frequent collaborators; Siyu An appears in middle-author positions on both papers. Output, impact, and activity cannot yet be scored for lack of bibliometric data. The name is resolved from public bibliographic data; institution and ORCID are unconfirmed.
Collaboration pattern
With 16 distinct coauthors and a recurring core of 12 frequent collaborators — among them Junnan Dong, Ruizhi Qiao, Xing Sun, Xiaoyu Shen, Ruifeng Xu, and Linhao Luo — Siyu An works within sizeable teams. Both reviewed 2026 preprints place them in a middle author position, not corresponding author, a contributor pattern within multi-member groups spanning NLP and graph machine learning.
Research keywords
Highlights
WFM: a Wiki Foundation Model over dense LLM-Wiki graphs
Latent neuro-symbolic disentangling of long-term agent memory
A consistent graph-plus-RAG research through-line
Graph neural networks and retrieval-augmented generation recur as keywords across both reviewed papers, matching the even 0.5/0.5 field split — a coherent agenda on memory-rich, retrieval-heavy agentic systems, currently at preprint stage.
Timeline
- 2026WFM: Wiki Foundation Model for Complex Agentic Reasoning appears on arXiv (middle-author position); Evolve's review notes the claimed 10.5x speedup and SOTA multi-hop QA gains alongside trust concerns.
- 2026Disentangling Long-Term Memory via Latent Neuro-Symbolic Reasoning appears on arXiv, introducing query-conditioned latent graphs for personalized agents with large reported benchmark gains.
- 2026Both preprints enter the Evolve catalog rated low reproducibility readiness, defining the current evidence base for this profile.
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 are computed from OpenAlex-based public bibliographic data; output, impact, and activity scores are absent because the underlying snapshot lacks sufficient data.
- The name is resolved from public bibliographic data; 'Siyu An' may be shared by multiple researchers, and institution, country, and ORCID are unconfirmed.
- This profile rests on only 2 arXiv preprints reviewed by Evolve (both 2026); peer-reviewed venue history is not represented in the data.
- Claims such as the '10.5x speedup' and 'SOTA QA gains' are as stated in the papers and flagged by Evolve review, not independently verified.