Jin Xu

Alibaba Group (China)

Unclaimed · AI-structured profileOpenAlex

Alibaba Group researcher in multimodal ML: audio-visual dialogue, synthetic data engines, and intent benchmarks

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: the catalog holds only two 2026 arXiv papers for this name, too thin to measure output. This axis would aggregate publication volume, venue quality, and authorship position to gauge sustained research production.

Impact—

Not yet scored: no citation counts or downstream-usage signals are available for these two recent preprints. This axis would measure citations, adoption of the released benchmarks and data engines, and influence on later audio-visual dialogue work.

Focus0/100

The radar scores focus at 0, the low end of the scale. The raw field distribution explains why: activity splits evenly (0.5/0.5) between two adjacent subfields — multimodal machine learning for speech and audio-visual interaction, and for speech and vision. Both catalog papers sit inside this audio-visual space, so the split reads as deliberate breadth around one problem area rather than scattered topic-hopping, though the sample is only two papers.

Collaboration70/100

Collaboration scores 70, backed by 32 distinct coauthors across the catalog footprint. Recurring names include Yunfei Chu (listed twice among frequent coauthors), Qi Chen, Xie Chen, Qiuqiang Kong, Yuxuan Wang, and Ziyang Ma. Both papers carry middle authorship with no corresponding role, consistent with contributing inside large industrial teams at Alibaba Group rather than small lab-led groups.

Activity—

Not yet scored: the profile covers only two 2026 preprints, so an annual publication cadence cannot be established. This axis would track output per year and gaps in the record to describe how consistently this researcher publishes over time.

Reproducibility78/100

Reproducibility is the strongest measured axis, with a mean of 78 across 2 papers. Per-paper readiness differs: ODU-Bench is rated high, consistent with an explicit benchmark artifact, while OmniVChat is rated medium, reflecting a heavier pipeline of synthetic-data engines, multi-agent generation, and reinforcement-learning training that is harder to re-run. The 78 comes directly from Evolve's repro-readiness reviews of both papers.

Research summary

Jin Xu is a researcher at Alibaba Group (China) working at the intersection of multimodal machine learning, speech, and audio-visual interaction. The Evolve catalog holds two 2026 arXiv papers for this name, both in middle-author positions and neither as corresponding author — a contributor role inside large industrial teams rather than a sole-lead one. The first, OmniVChat, builds a full stack for native audio-visual dialogue: a multi-agent synthetic-data engine, a rubric-based benchmark, and reinforcement learning with rubric rewards; Evolve notes gains transfer to real recordings, while flagging judge bias and remaining trade-offs. The second, ODU-Bench, formalizes demand detection and contextual user-intent inference in multimodal interaction as an explicit benchmark, with a solid pipeline but an LLM-judged metric carrying 60% of the score. Bibliometric signals align: 32 distinct coauthors, a collaboration score of 70, and a field distribution split evenly (0.5/0.5) between the speech-and-audio-visual and speech-and-vision subfields — breadth across two adjacent areas rather than a narrow niche. Reproducibility is the strongest measured axis, with a mean readiness of 78 across 2 papers (medium for OmniVChat, high for ODU-Bench). Output, impact, and activity lack sufficient data and are left unscored. This is an early-stage, name-resolved profile built from public bibliographic data.

Collaboration pattern

Jin Xu works in sizeable industrial teams: 32 distinct coauthors and a collaboration score of 70, with both papers in middle-author, non-corresponding positions. Recurring coauthor names include Yunfei Chu, Qi Chen, Xie Chen, Qiuqiang Kong, Yuxuan Wang, and Ziyang Ma — a mix spanning speech and audio-visual expertise. Based on a two-paper sample, the pattern suggests a specialist contributor embedded in Alibaba Group's audio-visual dialogue efforts rather than an independent group lead.

Research keywords

audio-visual dialogueomni-modal LLMssynthetic data generationmulti-agent data enginereinforcement learningrubric-based rewardbenchmarkmultimodal LLMdemand understandingintent inferenceaudio-visual interactionfalse trigger

Highlights

OmniVChat: a full pipeline for native audio-visual dialogue
Middle-author contribution to a system that synthesizes audio-visual dialogue data with a multi-agent engine, benchmarks it with rubric-based rewards, and trains omni-modal LLMs with reinforcement learning; Evolve finds the gains transfer to real recordings while noting judge bias and remaining trade-offs.
ODU-Bench: making user demand and intent measurable
Co-authored benchmark that turns demand detection and contextual user-intent inference in multimodal interaction into an explicit evaluation target, covering cases like false triggers; the pipeline is rated solid, though an LLM-judged metric carries 60% of the score, and repro-readiness is high.
An early but coherent audio-visual research footprint
Both 2026 catalog outputs sit in the same audio-visual multimodal space, with the field distribution split evenly (0.5/0.5) between speech-and-audio-visual and speech-and-vision work, 32 distinct coauthors, and above-average reproducibility (78) — a coherent, team-embedded start rather than a scattered record.

Timeline

2026Co-authored OmniVChat on arXiv: a synthetic-data engine, rubric benchmark, and reinforcement-learning pipeline for native audio-visual dialogue (middle authorship, non-corresponding).
2026Co-authored ODU-Bench on arXiv, establishing demand detection and contextual user-intent inference as an explicit multimodal benchmark (middle authorship, non-corresponding).
2026Both catalog outputs land in the same year within Alibaba Group's audio-visual multimodal work, making this a single-year, early-stage profile footprint with no earlier record in the catalog.

Papers in the Evolve catalog

Frequent co-authors

Qi Chen×2 Yunfei Chu×2 Haolin He×2 Wen Huang×2 Qiuqiang Kong×2 Xie Chen×2 Yunfei Chu×2 Yifan Yang×1 Zihan Liu×1 Yuxuan Wang×1 Ziyang Ma×1 Ruiyang Xu×1

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 and coauthor counts are computed from public bibliographic data only; citation, venue-tier, and funding records are not part of this snapshot.
'Jin Xu' is a common Chinese name; this profile is name-resolved from public bibliographic data and the Alibaba Group affiliation, so identity matching may be imperfect.
The Evolve catalog holds only two 2026 arXiv papers; output, impact, and activity axes are unscored for insufficient data, and the timeline is necessarily single-year.
Both papers are arXiv preprints (repro-readiness medium and high); peer-review status is not reflected in this profile.