NLP and text-analysis researcher: topic modeling, handwritten document recognition, explainable AI, and LLM evaluation
3
papers here
3
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
30
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—
The output axis is not scored: the completeness-checked publication counts needed to normalize productivity against career length and field norms are unavailable in this snapshot. It would measure annual and cumulative publication volume relative to discipline baselines.
Impact—
The impact axis is not scored: citation counts, citation-year distributions, and venue-level influence needed to benchmark this record against peers are absent from the snapshot. It would measure citation-based influence such as citations per work and field-normalized citation impact.
Focus0/100
The focus score of 0 reflects an even spread rather than a narrow profile: the provided field distribution splits exactly three ways — explainable AI for document image classification, NLP/computational social science, and NLP/computational linguistics at 33% each — and five topic clusters (topic modeling at 214, NLP techniques at 146, handwritten text recognition at 94, IR/search behavior at 83, advanced text analysis at 77) reinforce that breadth, ranging from historical document processing to LLM evaluation.
Collaboration51/100
The collaboration score of 51 rests on a moderate base: 12 distinct coauthors appear in the recent three-year window, and the frequency roster includes Emanuela Boros, Mickaël Coustaty, Senja Pollak, Matej Martinc, Ahmed Hamdi, and Adam Jatowt, pointing to stable, repeated partnerships rather than one-off teams. In the three 2026 Evolve-reviewed papers this researcher holds one middle and two last author positions, with no corresponding-author role recorded.
Activity100/100
The activity score of 100 is the standout signal: 67 works in the most recent three-year window, with year counts climbing from single digits in the 2000s to a 2021 peak of 56, then 39 (2022), 35 (2023), 20 (2024), 21 (2025), and 26 already in 2026. That pace — at least 20 works every year since 2019 — marks an exceptionally high, sustained publishing cadence.
Reproducibility55/100
Reproducibility scores 55, computed from Evolve's review of 3 papers with a mean readiness of 55. All three 2026 preprints carry a medium readiness rating. The reviews temper this: the multi-negative DPO paper reports strong ablations but no significance tests or released code, the LIME segmentation study is convincing but narrow (3 classes, one ResNet-50, a 60-image bias test), and the political-bias study's chat and downstream-fairness claims exceed its data.
Research summary
Antoine Doucet is a researcher at the University of Ljubljana (Slovenia) whose OpenAlex-indexed record spans 2002–2026, with the provided year counts summing to 400 works, an h-index of 30, and an i10-index of 82. Productivity has accelerated sharply: annual output climbed from single digits in the 2000s to a peak of 56 works in 2021, and the most recent three-year window holds 67 works — the basis for a maximum activity score. Topic modeling (214 topic assignments), natural language processing techniques (146), handwritten text recognition (94), information retrieval and search behavior (83), and advanced text analysis (77) lead the topic profile. The three Evolve-reviewed papers, all 2026 arXiv preprints, mirror an even three-way field split — explainable AI for document image classification, NLP/computational social science, and NLP/computational linguistics at 33% each — covering political-bias evaluation of LLMs via the Political Compass Test (middle authorship), segmentation choices for LIME explanations of document classifiers (last author), and multi-negative DPO for entity linking in historical newspapers (last author). Evolve's verdicts credit methodological rigor while flagging narrow evaluation scopes and, in two papers, missing significance tests or code. Recent collaboration spans 12 distinct coauthors, with Emanuela Boros, Mickaël Coustaty, Senja Pollak, Matej Martinc, and Adam Jatowt among recurring partners.
Collaboration pattern
Collaboration follows a steady small-team pattern: 12 distinct coauthors appear in the recent three-year window, and the recurring roster includes Emanuela Boros, Mickaël Coustaty, Quentin Telnoff, Robin Jarry, Fabrice Crohas, Luka Debevc, Nishan Chatterjee, Senja Pollak, Matej Martinc, Tien Nam Nguyen, Ahmed Hamdi, and Adam Jatowt. In the 2026 Evolve-reviewed work, positions are one middle and two last authorships, with no corresponding-author role recorded, suggesting shared leadership within long-running partner groups.
Research keywords
Highlights
Mapping LLM political bias across prompts, personas, and languages
When superpixel segmentation breaks LIME explanations on documents
Multi-negative DPO for linking entities in historical newspapers
Timeline
- 2002Earliest works in the provided record appear: 3 indexed publications in 2002, followed by 2 in 2003.
- 2010Output spikes to 15 indexed works in 2010, the first double-digit year after single-digit counts through the 2000s.
- 2013An earlier high-water mark: 22 indexed works in 2013, before the steeper climb of 2019–2021.
- 2021The provided counts peak at 56 indexed works in 2021 — the career high in this snapshot.
- 2026Three arXiv preprints enter the Evolve catalog: LLM political-bias evaluation, LIME segmentation analysis for document explainability, and multi-negative DPO for historical entity linking.
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.
- Bibliometrics come from OpenAlex only (identity: openalex); h-index, i10-index, and per-year counts reflect that snapshot alone.
- Output and impact dimensions are unscored because citation-level and completeness-checked data are not present in the snapshot.
- The three reviewed papers are 2026 arXiv preprints; Evolve verdicts and reproducibility ratings cover only these works.
- Identity is name-resolved from public bibliographic data (OpenAlex/ORCID 0000-0001-6160-3356); affiliation is shown as provided in the input.