Benchmark-driven NLP for the Semantic Web: ontology learning, embeddings, and LLM-scale evaluation at TIB, Germany
3
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
16
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-17) 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 underlying data is insufficient for this axis. It would measure total publication volume and its consistency over time against field baselines; no normalized output score has been derived from this record.
Impact—
Not yet scored: citation evidence in this record is insufficient. The axis would measure influence via aggregate citations, field-normalized citation rates, and citing-work reach; these aggregates are not yet available, so no estimate is offered.
Focus—
Not yet scored: the data needed to quantify topical concentration is insufficient. This axis would measure how evenly output spreads across topics versus concentrating in a few themes; no normalized focus score has been computed for this profile yet.
Collaboration32/100
The collaboration axis scores 32/100, drawn from a compact network: only 4 distinct co-authors appear in the recent three-year OpenAlex window. Recurring partners are Hamed Babaei Giglou, Sören Auer, Peio Popov, and Mahsa Sanaei, and on all three cataloged 2026 papers D'Souza sits in a middle or last author position, consistent with senior supervisory authorship inside a small stable team rather than consortium-scale collaboration.
Activity100/100
The activity axis scores 100/100. The record shows 96 works in the recent three-year window, and yearly output has climbed from 1-6 papers during 2012-2015 to 25 in 2024, 37 in 2025, and 34 already in 2026 — a steep, sustained acceleration that makes the last three years the most productive stretch of the recorded career.
Reproducibility—
Not yet scored: artifact-level evidence is insufficient for this axis. It would measure whether publications release reusable code, data, and benchmarks; the three cataloged papers carry only a coarse qualitative reproducibility flag rather than a computed score.
Research summary
Jennifer D'Souza is a researcher at the Technische Informationsbibliothek (TIB) in Germany whose work sits at the intersection of natural language processing and the Semantic Web. Her OpenAlex record shows an h-index of 16, an i10-index of 23, and a sharply accelerating output curve: yearly counts rise from 1-6 papers across 2012-2015 to 25 in 2024, 37 in 2025, and 34 already in 2026, with 96 works in the most recent three-year window. Topic Modeling (117 topic-assigned works) dominates, followed by Semantic Web and Ontologies (66), Biomedical Text Mining and Ontologies (55), Advanced Text Analysis Techniques (37), and Natural Language Processing Techniques (29). The three Evolve-cataloged 2026 arXiv papers capture her current agenda: stress-testing whether general sentence embeddings encode ontological reasoning via the AVA benchmark, where 25+ embeddings fail at ontology level while fine-tuning nearly solves it yet transfers poorly; fusing string-based, knowledge-graph-embedding, and LLM ontology aligners through voting (OntoAligner-Ensemble); and a controlled 13-model study showing LLM scale mainly lifts ontology-learning precision, non-monotonically. She publishes within a small, stable co-author group — 4 distinct recent co-authors — including Hamed Babaei Giglou, Sören Auer, Peio Popov, and Mahsa Sanaei, often in middle or last author position. Evolve reviews credit sound benchmark design while flagging thin test sets and missing significance tests.
Collaboration pattern
D'Souza's collaboration footprint is compact: the OpenAlex record lists 4 distinct recent co-authors and the collaboration axis scores 32/100. Her recurring partners are Hamed Babaei Giglou, Sören Auer, Peio Popov, and Mahsa Sanaei. On all three Evolve-cataloged 2026 papers she takes a middle or last author position rather than first or corresponding, a pattern consistent with senior, supervisory authorship within a small stable group rather than large multi-institution consortia.
Research keywords
Highlights
25+ sentence embeddings fail ontology-level reasoning on the AVA benchmark
OntoAligner-Ensemble fuses string, KGE and LLM aligners by voting
Controlled 13-model study probes when LLM scale helps ontology learning
Timeline
- 2026Publishes the AVA-benchmark study showing that general NLP embeddings fail ontology-level discrimination while fine-tuning nearly solves it.
- 2026Releases OntoAligner-Ensemble, a voting-based fusion of string-based, knowledge-graph-embedding and LLM ontology aligners.
- 2026Reports a controlled 13-model study finding that LLM scale mainly raises ontology-learning precision, non-monotonically.
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.
- Bibliometric indicators (h-index, i10-index, yearly counts, topics) come from OpenAlex and may undercount non-indexed venues or lag recent publications.
- Paper-level analysis rests on only three 2026 arXiv preprints in the Evolve catalog; peer-reviewed versions may differ.
- Output, impact, focus, and reproducibility axes are unscored because the underlying data is insufficient.
- Identity is resolved from public OpenAlex records against the supplied TIB affiliation; same-name collisions cannot be fully excluded.