Le Thien Phuc Nguyen

Research profile

Research summary

Le Thien Phuc Nguyen is a researcher profiled from public OpenAlex bibliographic data; the snapshot carries no institution, country or ORCID. The indexed record shows two phases. An earlier sustained stretch (2012–2018) concentrates on Human Mobility and Location-Based Analysis (11 topic-assigned works) and Context-Aware Activity Recognition Systems (9), with smaller clusters in Domain Adaptation and Few-Shot Learning (5) and privacy-related topics (4 and 4). After a gap, publishing resumes in 2025–2026 (3 works in the last three years) around multimodal vision-language systems. The 2026 arXiv output includes QwenVLConnector, a first-author medical vision-language chatbot on Qwen2.5-VL targeting fine-grained clinical perception and radiology report generation on FLARE-2D tasks — Evolve judges its gains plausible but attribution-confounded, with regression and multi-label performance worsening — and DocHop, a carefully verified but fully synthetic, template-bound benchmark for out-of-domain multi-hop chart reasoning, rated high for reproducibility readiness. OpenAlex records an h-index of 13, an i10-index of 14, 12 distinct coauthors and a thematic-focus score of 100, the profile's strongest dimension. Output and impact axes remain unscored because per-paper citation detail is absent, and the year counts include anomalous 1976–1978 entries suggesting possible homonym contamination.

Research keywords

Highlights

• First-author medical VLM chatbot built on Qwen2.5-VL

  QwenVLConnector (2026, arXiv) proposes a dense connector for fine-grained clinical perception and radiology report generation on FLARE-2D tasks over Qwen2.5-VL. Evolve's verdict: the gains are plausible but attribution is confounded, and regression and multi-label performance worsen; reproducibility readiness is medium.

• DocHop: a verified synthetic benchmark for out-of-domain multi-hop chart reasoning

  This 2026 arXiv benchmark isolates chart-context multi-hop reasoning in information-dense documents. Evolve calls it carefully verified and well-engineered, but fully synthetic and template-bound, limiting ecological validity; its reproducibility readiness is rated high.

• Earlier record concentrated in human mobility and activity recognition

  Beyond the two 2026 releases, OpenAlex topic counts show 11 works in Human Mobility and Location-Based Analysis and 9 in Context-Aware Activity Recognition Systems, with side clusters in domain adaptation/few-shot learning (5) and privacy-preserving technologies (4 and 4).

Timeline

1. 2012Earliest clustered works appear in the OpenAlex record, with 4 publications indexed in 2012.
2. 2014Peak indexed output: 6 publications, within an active 2012–2015 stretch (4, 3, 6, 4).
3. 2015–2018Cadence tapers: 4 works in 2015, then 1 per year in 2016, 2017 and 2018.
4. 2025Renewed indexed activity after the gap since 2018: 1 publication recorded.
5. 2026Both Evolve-cataloged arXiv papers appear — first-author QwenVLConnector and DocHop (middle author).

Papers in the Evolve catalog

• 2026-09-19first author

  QwenVLConnector: A Fast, Unified Medical VLM Chatbot for Fine-Grained Clinical Perception and Text Generation

• 2026-09-02

  DocHop: Benchmarking Out-of-domain Multi-hop Reasoning in Information-Dense Documents

Frequent co-authors