Alibaba Group researcher bridging LLM emotion companions, agentic RL horizons, and efficient long-context attention
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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 cannot be scored yet: beyond the three 2026 preprints reviewed here, no verified publication-count, venue, or authorship-role series exists. This axis would measure cumulative paper volume, venue tier, and first/corresponding-author share over time.
Impact—
Impact is unscored: the input data contains no citation counts, h-index, or adoption signals, and all reviewed papers are 2026 preprints. This axis would measure citations, downstream uptake, and real-world use of the methods.
Focus0/100
The focus score is 0/100 because the field distribution is perfectly flat: human-computer interaction / affective computing, efficient long-context attention for LLMs, and LLM agents and reinforcement learning each hold exactly 0.333 of the three reviewed papers. With one paper per area and no dominant field, no concentration signal exists; the axis measures how evenly or narrowly output concentrates, and here it registers maximal spread.
Collaboration59/100
Collaboration scores 59/100 on 18 distinct coauthors across the reviewed work, with a recurring core of 12 frequent collaborators including Kexin Quan, Zijian Ding, Jiaye Yong, Qinshi Zhang, Jessie Chin, and Jie Zhou. The pattern suggests stable small-team work rather than wide rotating consortia; notably, Wang sits in a middle author position and is not corresponding author on any of the three reviewed papers.
Activity—
Activity is unscored: the data provides no multi-year publication cadence, only three 2026-dated preprints. This axis would measure output rate over time — papers per year, submission rhythm, and continuity of research engagement across calendar years.
Reproducibility43/100
Reproducibility averages 43/100 across three scored papers. The emotion-companion field study is rated low — a 14-day, 19-user, 1,093-session deployment without a control group resists replication. The two ML papers rate medium: Elastic Horizon is well-ablated but its 7B margins sit near seed noise and its 14B result is single-seed, while CEDAR claims ~3× 128K speedup from single-model results without variance data.
Research summary
Dong Wang is a researcher at Alibaba Group (China) whose reviewed output spans three areas in near-equal measure: human-computer interaction / affective computing, efficient long-context attention for large language models, and LLM agents with reinforcement learning — each field holding exactly one of the three 2026 arXiv preprints in Evolve's catalog, with Wang in a middle author position and never the corresponding author. The HCI strand contributes a 14-day mixed-methods field study of an LLM emotion companion (19 users, 1,093 sessions), judged compelling but weakened by the absence of a control group. The ML strand pairs CEDAR, an error-bounded residual-routing sparse-attention method claiming roughly 3× 128K prefill speedup, with Elastic Horizon, a closed-loop controller that discovers the effective interaction frontier for agentic RL using P90 success lengths plus an EMA. The radar scores collaboration 59/100 on 18 distinct coauthors, reproducibility 43/100 across three scored papers, and focus 0/100, reflecting the perfectly even three-way field split; output, impact and activity axes remain unscored for lack of data. Treat this as a name-resolved profile built solely from the reviewed papers and their bibliometric snapshot.
Collaboration pattern
Across the three reviewed 2026 preprints, Dong Wang works in team settings from a middle author position, never as corresponding author. The 18 distinct coauthors include a recurring core — Kexin Quan, Zijian Ding, Jiaye Yong, Qinshi Zhang, Jessie Chin, Siyu Li, Jie Zhou, Wei Li, Yang Xu — suggesting stable internal collaboration groups spanning both the HCI/affective computing and the LLM systems/RL strands.
Research keywords
Highlights
14-day field study of an LLM emotion companion (19 users, 1,093 sessions)
Elastic Horizon: closed-loop control of the agentic RL interaction frontier
CEDAR: error-bounded residual routing claiming ~3× 128K prefill speedup
Timeline
- 2026HCI/affective computing strand: 14-day mixed-methods field study of an LLM emotion companion (19 users, 1,093 sessions).
- 2026LLM agents strand: proposes Elastic Horizon, a closed-loop horizon controller for agentic reinforcement learning.
- 2026ML systems strand: proposes CEDAR, error-bounded residual routing for efficient long-context attention, claiming ~3× 128K prefill speedup.
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.
- Radar scores are derived from a limited public bibliometric snapshot; citation, venue and output metrics are absent, so output, impact and activity axes are unscored.
- All three reviewed papers are 2026 arXiv preprints; no peer-reviewed venue record appears in the catalog.
- "Dong Wang" is a common name and no ORCID is provided; this profile is name-resolved from public bibliographic data and identities may merge or split.
- Author-position metadata covers only the three reviewed papers; broader publication and authorship history is unknown.