Researcher at the NLP–ML interface: LLM-agent long-term memory, graph foundation models, retrieval-augmented generation
3
papers here
3
reviewed by Evolve
1
first-author
0
corresponding
—
works · OpenAlex
11
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—
Publication-volume metrics — total outputs, venue record, and output trend — are not available in the current snapshot, so this axis cannot be scored yet. It would measure annual research output and venue profile once a fuller bibliographic record is linked.
Impact—
Citation counts, h-index, and downstream uptake are absent from the data, so impact cannot be estimated. This axis would measure how often and how widely the researcher's papers are cited and built upon by later work.
Focus0/100
The focus score is 0, derived from an even 50/50 field distribution between natural language processing (LLM agents, long-term memory and personalization) and machine learning (graph representation learning and retrieval-augmented generation). The two strands are technically adjacent — both catalog papers share graph neural networks and RAG — but a two-paper record is too thin for a stable specialization reading.
Collaboration63/100
With a collaboration score of 56, the record shows a moderately networked pattern: 16 distinct coauthors are linked, and a recurring circle of 12 frequent partners — including Ruifeng Xu, Linhao Luo, Xiaoyu Shen, Xing Sun, and Ruizhi Qiao — surrounds the reviewed work. Both catalog papers are multi-authored, with first-authorship on WFM and a middle position on the memory paper.
Activity100/100
Only two 2026 papers are on record, which is too few to compute a recent-activity trend, so this axis is left unscored. It would measure publishing cadence and year-over-year continuity once more publication years become available.
Reproducibility32/100
Reproducibility scores 20, the mean across 2 reviewed papers, both rated low readiness. Evolve's reviews record concrete gaps: the latent neuro-symbolic memory paper lacks significance tests, a stated judge, and code, while WFM claims a 10.5x speedup and SOTA QA gains but carries a leaked 'planning values' note and absent statistics. Released code and reported variance would materially raise this axis.
Research summary
Junnan Dong (ORCID 0000-0003-2117-6083) is an early-stage profile name-resolved from public bibliographic data, working at the intersection of natural language processing and machine learning. The Evolve catalog holds two 2026 arXiv preprints that split evenly — 50/50 in field distribution — between LLM-agent research on long-term memory and personalization, and graph representation learning with retrieval-augmented generation. The first-authored WFM: Wiki Foundation Model for Complex Agentic Reasoning builds a graph foundation model over dense LLM-Wiki graphs and claims a 10.5x speedup with state-of-the-art QA gains; Evolve's review flags a leaked 'planning values' note and absent statistics as trust concerns. The second paper disentangles long-term memory via latent neuro-symbolic reasoning: its query-conditioned latent graphs show large personalization-benchmark gains, but evaluation lacks significance tests, a stated judge, and code. Both papers carry low reproducibility readiness, yielding a reproducibility score of 20 over 2 reviewed papers. Collaboration scores 56, with 16 distinct coauthors and a recurring circle including Ruifeng Xu, Linhao Luo, and Xiaoyu Shen. Institution, country, and citation-impact data are not available in the current snapshot.
Collaboration pattern
The record links 16 distinct coauthors, with a stable recurring circle of 12 frequent partners including Ruifeng Xu, Linhao Luo, Xiaoyu Shen, Xing Sun, Ruizhi Qiao, and Keyu Chen. Both reviewed papers are multi-authored team efforts — first-authorship on WFM and a middle position on the latent neuro-symbolic memory paper — suggesting an active contributor within larger LLM-agent and graph-learning groups rather than a solo line of work.
Research keywords
Highlights
First-authored WFM: a Wiki foundation model for agentic reasoning
Query-conditioned latent graphs for long-term memory (LGM)
A 50/50 dual-track across LLM agents and graph learning
The field distribution splits evenly between LLM-agent memory work and graph representation learning, yet both catalog papers share a common toolbox of graph neural networks and retrieval-augmented generation.
Timeline
- 2026First-authored WFM: Wiki Foundation Model for Complex Agentic Reasoning posted to arXiv, claiming a 10.5x speedup and state-of-the-art QA gains.
- 2026Co-authored work on disentangling long-term memory via latent neuro-symbolic reasoning posted to arXiv, reporting large personalization-benchmark gains.
- 2026Both catalog outputs land in the same year, splitting evenly between LLM-agent memory research and graph-based retrieval-augmented generation.
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 dimensions are computed from OpenAlex-based public bibliographic data only; citation and venue metadata are absent from this snapshot.
- Institution and country are unresolved; the profile is name-resolved from ORCID 0000-0003-2117-6083 and public bibliographic records, so identity confidence is limited for a common name.
- Qualitative verdicts rest on 2 Evolve-reviewed 2026 arXiv preprints; no peer-reviewed venue record is captured.
- The reproducibility score (20) covers only the 2 reviewed papers, both rated low readiness.