Lemen Chao

Research profile

Research summary

Lemen Chao is a researcher at Renmin University of China working in explainable artificial intelligence, with a 2026 arXiv record showing two first-author papers that bracket the field's methodological and human-centered halves. The first, ADORE, proposes a unified framework for global and local interpretability using adaptive derivative-ordered random explanation, combining finite differences, Hessian/Jacobian approximation, and randomized SVD; Evolve's review finds its efficiency gains credible while flagging flawed adaptivity logic and missing significance tests. The second converts SHAP attributions into ABT-structured data storytelling, using large language models to explain AI decisions to non-experts without exposing sensitive data or model details; the reported survey gains are significant but rest on single-case evidence with small, uncorrected effects. Bibliometric context is thin but consistent: an h-index of 4, an i10-index of 2, 4 distinct coauthors (Ming Lei, Anran Fang, Zixuan Yang, Mingran Sun), and an even 50/50 field split between machine learning/XAI and human-centered XAI. OpenAlex counts-by-year show early output from 2009 (peaking at 3 works per year in 2010 and 2011), intermittent single works through 2020, and 4 works in 2026; the recent-3-year count is 4. Evolve's repro-readiness mean across the 2 reviewed papers is 55 (both rated medium). Output and impact axes remain unscored for lack of citation-level data.

Research keywords

Highlights

• ADORE: unifying global and local interpretability through derivatives

  First-author 2026 arXiv paper unifying global and local explainable AI via adaptive derivative-ordered random explanation, combining finite differences, Hessian/Jacobian approximation, and randomized SVD. Evolve's review finds the efficiency gains credible but flags flawed adaptivity logic and no significance tests; repro-readiness is rated medium.

• Data storytelling for non-experts: SHAP to ABT narratives

  First-author 2026 arXiv framework that turns SHAP attributions into ABT-structured stories, using large language models to explain AI decisions to non-experts without revealing sensitive data or model details. Evolve notes significant survey gains, tempered by single-case evidence and small, uncorrected effects; repro-readiness is medium.

• 2026 return to publishing with four works

  OpenAlex counts-by-year record 4 works in 2026 — the highest annual total in the record, matching the 2010–2011 peak — including the two first-author explainable-AI preprints reviewed by Evolve; the recent-3-year count is 4 works.

Timeline

1. 2009Earliest publication year in the OpenAlex record, with 1 work.
2. 2010Annual output peaks at 3 works, matched again in 2011.
3. 2014Closes an early sustained stretch of 1–3 works per year (2009–2014).
4. 2020Last year of an intermittent stretch with single works in 2017–2020.
5. 2026Returns with 4 works, including two first-author explainable-AI arXiv papers in the Evolve catalog.

Papers in the Evolve catalog

• 2026-09-15first author

  A unified framework for global and local interpretability using adaptive derivative-ordered random explanation

• 2026-09-14first author

  Data storytelling meets interpretable machine learning: Decoding AI decisions for non-experts without revealing sensitive data and model details

Frequent co-authors