Pith. sign in

REVIEW 3 major objections 7 minor 1 cited by

Retrieval-Augmented Generation of Ontologies from Relational Databases

T0 review · 3 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that retrieval-augmented iterative generation can turn relational database schemas into rich, standards-compliant OWL ontologies that outperform baseline and non-iterative methods, with mean competency-question scores…

desk verdict RIGOR is a genuinely new and useful pipeline for relational-to-OWL conversion, but the paper's headline comparison is undermined by a judge-LLM that can answer the test questions from the database schema alone. read the letter →

arxiv 2506.01232 v1 pith:MV5RPMGA submitted 2025-06-02 cs.DB cs.AI

Mojtaba Nayyeri , Athish A Yogi , Nadeen Fathallah , Ratan Bahadur Thapa , Hans-Michael Tautenhahn , Anton Schnurpel , Steffen Staab This is my paper
classification cs.DBcs.AI
keywords RelationaldatabaseOntologyLargelanguagemodelsRetrieval-augmentedgenerationOWL2DLKnowledgegraphCompetencyquestionsLLM-as-judge
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

RIGOR is an LLM-driven pipeline that turns a relational database schema into an OWL 2 DL ontology with minimal human effort. The paper claims that by iterating table-by-table along foreign keys and retrieving relevant context from the schema documentation, an external ontology repository, and the ontology built so far, the pipeline produces ontologies that are logically consistent, syntactically valid, and semantically aligned with the database. Evaluated on a hospital liver-cancer registry and the public ICU database, RIGOR's outputs score above 4.5 out of 5 on six quality dimensions and show few critical modeling pitfalls, beating both a schema-only baseline and a single-pass non-iterative variant. The payoff, if the claim holds, is that a traditionally labor-intensive knowledge-engineering step can be largely automated.

What carries the argument

The load-bearing mechanism is the provenance-tagged delta ontology fragment produced for each table and the Gen-LLM/Judge-LLM refinement loop. Embedding-based retrieval using Faiss and cosine similarity pulls in the table's schema text, its documentation, matching concepts from external domain ontologies, and already-built core content; a prompt template then constrains the generator with rules such as one domain and range per property, no generic 'is' property, and universal or existential restrictions for table attributes. The judge model checks each fragment for coherence with the core, schema alignment, syntactic and logical validity, and naming clarity before the fragment is merged, and the whole cycle repeats by following foreign-key links until every table is covered.

What would settle it

Re-run the evaluation with the judge prompt stripped of the database schema, keeping only the ontology fragment and the competency questions; if RIGOR's advantage over the non-iterative baseline vanishes, the headline result is an artifact of schema leakage.

Watch

Extended reading notes

Core claim

The central discovery claimed is that a retrieval-augmented, iterative 'generate then judge' loop makes large language models reliable ontology engineers for relational data. RIGOR processes each table as a delta ontology fragment with provenance annotations, refines the fragment with a judge LLM, and merges it into a growing core ontology, so the final artifact satisfies schema coverage, semantic alignment, provenance, and external alignment requirements. The authors report that all RIGOR-generated ontologies parse as OWL 2 DL, pass HermiT consistency checks, and receive mean competency-question scores above 4.5/5, while baseline and non-iterative methods frequently produced invalid syntax. The paper presents this as evidence that retrieval-guided LLMs can generate semantically rich, standards-compliant ontologies that faithfully mirror complex relational schemas.

Load-bearing premise

The quality comparison rests on a judge LLM whose prompt includes the database schema, so the high scores may reflect schema informativeness rather than ontology content.

Editorial extensions

If this is right

  • Relational-to-ontology mapping can be automated for biomedical databases without a domain expert writing axioms by hand.
  • Generated ontologies satisfy schema coverage, semantic alignment, provenance annotation, and external alignment, going beyond the structural mirror produced by W3C Direct Mapping and R2RML.
  • RIGOR outperforms both a schema-only baseline and a single-pass non-iterative approach on competency-question quality, across three different generator LLMs and two databases.
  • All RIGOR-generated ontologies pass OWL 2 DL syntax checks and HermiT consistency, and OOPS! finds no critical modeling pitfalls, only minor refinements such as multiple domains or ranges.
  • The same fragments carry provenance metadata pointing to their source tables, which supports traceability and incremental updates.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the judge prompt includes the database schema, the reported 4.5+ scores may partly measure how much the schema alone reveals the answers to the competency questions; a blinded judge that sees only the ontology fragment would isolate the ontology's own contribution.
  • The pipeline should transfer to other domains whenever a comparable external ontology repository exists, so the design is not inherently tied to medicine.
  • The delta-fragment structure could be reused to emit R2RML mappings directly, turning the generated ontology into a queryable virtual knowledge graph over the live database.
  • The iterative, table-by-table design also suggests a natural maintenance mode: when a schema column changes, only the affected table's fragment needs regeneration.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 7 minor

Summary. The paper introduces RIGOR, an iterative retrieval-augmented generation pipeline that transforms relational database schemas into OWL 2 DL ontologies. The pipeline traverses tables by foreign-key order, retrieves context from schema documentation, an external ontology repository, and a growing core ontology, then prompts a generative LLM to produce provenance-tagged delta ontology fragments in Manchester syntax. Fragments are reviewed and refined by a judge LLM before being merged. The authors evaluate RIGOR on a liver-cancer registry and the PhysioNet ICU database, comparing it against a schema-only baseline and a non-iterative approach with three different generative LLMs. They report that RIGOR achieves higher competency-question scores, syntactic validity, logical consistency, and few critical modeling pitfalls, with reduced manual effort.

Significance. The problem addressed is important and timely: converting relational databases into semantically enriched ontologies with minimal human effort. The paper provides a concrete, reproducible system with released code, and the machine-checked claims about RIGOR's own outputs (syntax, HermiT consistency, OOPS! pitfalls) are a strength. If the comparative evaluation were robust, RIGOR would be a valuable contribution to ontology learning and semantic data integration. However, the head-to-head comparison depends on a judge-LLM that receives the database schema in its prompt, so the advertised quality gains are not yet convincingly established.

major comments (3)
  1. [§4.5 (Strategy 6), Appendix Fig. 4] The Judge-LLM prompt includes the Database Schema alongside the ontology fragment and the competency questions, and the CQs themselves are generated from the same schema (Appendix Fig. 3). The judge can therefore answer most questions from the schema alone, so the 0–5 scores do not isolate the ontology's contribution. This concern is concrete: Table 2 shows that non-iterative outputs, which Section 4.6 states 'often failed to produce valid ontologies', still receive 3.3–3.7, not far below RIGOR's 4.2–4.6. I request a control experiment (e.g., scoring schema-only vs. ontology-only, or evaluating the judge on a gold standard) or a human-judge calibration sample to verify that the judge measures ontology content rather than schema informativeness.
  2. [§4.6] Strategies 1–5 (syntax, consistency, OOPS!, structural metrics, semantic coverage) are reported only for RIGOR-generated outputs, so the entire RIGOR-vs-baseline comparison rests on Strategy 6. Given the concern about Strategy 6, the comparative claim is currently unsupported by independent evidence. Please apply at least the syntax and consistency checks to baseline and non-iterative outputs, or report their failure rates, and ideally extend the semantic coverage analysis to all methods rather than a single table (chemotherapy only).
  3. [Abstract and §5] The claim 'achieving mean CQ scores above 4.5/5' is not supported by Table 2: Mistral-3.1 under RIGOR scores 4.2±0.22 on the real-world database and 4.3±0.15 on ICU. Either correct the claim or report per-method, per-dimension scores that support it. This is a factual discrepancy in the conclusion, not merely a stylistic issue.
minor comments (7)
  1. [§3.1, Eq. (3)] The displayed definition of NDoc_u has a line break and an equation number placed inside the set-builder notation, making the formula difficult to read; please reformat.
  2. [§2.3] The phrase 'advanced hybrid recursive RAG' is not defined; the taxonomy introduced earlier in the section is not explicitly linked to the concrete choices made in Section 3.
  3. [References] Reference [7] is duplicated (appears as refs 6 and 7), and the citation '[49,46,49,39]' repeats [49]; please clean up the reference list and citations.
  4. [Appendix, Tables 6–7] The captions say the comparison is across Baseline, Non-Iterative, and RIGOR methods, but the tables only list rows per gen-LLM with no method split; clarify how these aggregates map to Table 2.
  5. [§4.3 vs. §3.6] The Mistral model identifier is inconsistent: 'Mistral-Small-3.1-24B-Instruct-2503' in Section 4.3 versus 'Mistral-Small-24B-Instruct-2501' in Section 3.6.
  6. [§4.5, Strategy 5] Strategy 5 uses all-MiniLM-L6-v2 for embedding, whereas Section 3.2 specifies all-mpnet-base-v2 for retrieval; please state whether this difference is intentional and how it affects the semantic coverage result.
  7. [§3.6] The statement that CQs were 'verified by a domain expert' lacks details on the verification protocol, the number of experts involved, and inter-annotator agreement; please provide this information or soften the claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: RIGOR's generation and evaluation chain is self-contained, with no load-bearing step reducing to its own inputs by construction.

full rationale

The paper's claimed derivation is an engineering pipeline rather than an analytical reduction: it takes an RDB schema, documentation, and external ontologies, retrieves relevant context, prompts a Gen-LLM to produce delta-ontology fragments, refines them via a judge-LLM, and merges them iteratively. The central comparative claim (RIGOR outperforms baseline and non-iterative variants) rests on Strategy 6, the Judge-LLM CQ scores. Although the judge prompt in Appendix 6.2, Fig. 4 includes the database schema, this is an evaluation-design confound rather than circularity: the score is not defined as a deterministic function of the schema, and the same schema is present for all compared methods. The judge still gave baseline outputs mean scores near 1.3-1.6 and RIGOR outputs 4.2-4.6, showing that the judge does distinguish the ontology fragments rather than simply rewarding schema information. The CQs are generated from the table schema, but that makes them standard requirement-elicitation artifacts; the ontology is not fitted to the CQ scores, and no equation equates the output to the input by construction. Strategies 1-5 (Protege parsing, HermiT consistency, OOPS! pitfalls, structural counts, and embedding-based column coverage) provide independent external grounding for the RIGOR outputs, even though those strategies were not applied to the invalid baseline outputs. Co-authored references (e.g., Volz et al., Xiong & Staab, Fathallah et al.) appear only as related-work background and are not load-bearing for the method or its evaluation. A judge protocol that hid the schema or added human evaluation of all methods would strengthen the paper, but that is a correctness and validity concern, not circularity in the derivation chain.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim relies on the LLM pipeline components behaving as specified and on the evaluation instruments being valid. The most consequential free parameter is the similarity threshold used to compute the single reported coverage number; the most consequential assumptions are that the external ontologies and the GPT-4o-generated documentation are correct and that the retrieval and judging LLMs work as intended.

free parameters (3)
  • Semantic coverage similarity threshold = 0.55
    In Section 4.5 Strategy 5, the threshold for matching column names to ontology class names is 'empirically tuned'; the coverage rate 72.4% depends directly on it.
  • Top-k retrieval count = not reported
    Equations 1-4 retrieve the top-k most similar elements from each source, but k is never specified; it controls context size and downstream output.
  • Gen-LLM decoding parameters = temperature 0.7, max_new_tokens 2000 (Appendix 6.3)
    Generation and scoring are stochastic; the reported 0-5 scores and their standard deviations depend on these hand-chosen decoder settings.
assumptions (5)
  • domain assumption OWL 2 DL and Manchester syntax are the appropriate target representation for the generated ontology fragments
    The pipeline and consistency checks (HermiT, Protégé) assume OWL 2 DL semantics; if the target were a lighter profile, some pitfalls would be irrelevant.
  • domain assumption The four BioPortal ontologies (Cell Ontology, DO, ICD-10, ONST) are authoritative and semantically correct for the medical domains in the databases
    Used as external grounding in Section 4.2 without verification of their suitability for the specific hospital registry and ICU schema.
  • domain assumption The embedding model all-mpnet-base-v2 and Faiss cosine retrieval surface semantically relevant concepts from all three knowledge sources
    Section 3.2 assumes that lexical-embedding similarity is sufficient to retrieve useful ontology alignment targets; no retrieval quality evaluation is reported.
  • domain assumption The Gen-LLM and Judge-LLM produce valid Manchester Syntax and meaningful feedback respectively
    The pipeline's correctness depends on the LLMs following the prompt format exactly (Sections 3.3-3.4); failures would produce fragments that cannot be parsed or merged.
  • domain assumption GPT-4o-generated natural language documentation, reviewed by medical professionals, accurately describes the database columns and tables
    Section 4.3 states the descriptions were reviewed by medical professionals, but the review method and the resulting accuracy are not reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Retrieval-Augmented Generation of Ontologies from Relational Databases." pith.science (2026). https://pith.science/paper/MV5RPMGA

@misc{pith2026250601232,
  author       = {Pith},
  title        = {Pith review of: Retrieval-Augmented Generation of Ontologies from Relational Databases},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MV5RPMGA}},
  note         = {Machine review of arXiv:2506.01232}
}
read the original abstract

Transforming relational databases into knowledge graphs with enriched ontologies enhances semantic interoperability and unlocks advanced graph-based learning and reasoning over data. However, previous approaches either demand significant manual effort to derive an ontology from a database schema or produce only a basic ontology. We present RIGOR, Retrieval-augmented Iterative Generation of RDB Ontologies, an LLM-driven approach that turns relational schemas into rich OWL ontologies with minimal human effort. RIGOR combines three sources via RAG, the database schema and its documentation, a repository of domain ontologies, and a growing core ontology, to prompt a generative LLM for producing successive, provenance-tagged delta ontology fragments. Each fragment is refined by a judge-LLM before being merged into the core ontology, and the process iterates table-by-table following foreign key constraints until coverage is complete. Applied to real-world databases, our approach outputs ontologies that score highly on standard quality dimensions such as accuracy, completeness, conciseness, adaptability, clarity, and consistency, while substantially reducing manual effort.

Figures

Figures reproduced from arXiv: 2506.01232 by the authors.

Figure 1
Figure 1. Overview of our proposed iterative pipeline: (a) traverse tables following foreign [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 1
Figure 1. RIGOR builds the ontology by iteratively traversing the tables of the [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Prompt template used for ontology generation. Fixed delimiters such as [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The evaluation dimensions included: (a)Accuracy: Correctness of do￾main representation and alignment with CQs. (b) Completeness: Coverage of essential concepts and relationships. (c) Conciseness: Absence of redundancy in the generated ontology. (d) Adaptability: Extens…
Figure 3
Figure 3. Figure 3: (a) The CQs Generation prompt: The prompt used to guide the LLM in generating CQs relevant to each database table. It was refined through empirical tuning of prompt engineering techniques to encourage structured, ontology-aligned outputs. Fixed delimiters such as **1. …
Figure 4
Figure 4. Figure 4: (b) Ontology quality evaluation through CQ performance prompt: The prompt used to evaluate the quality of ontology fragments with respect to specific metrics (e.g., accuracy, completeness). The Judge-LLM receives the CQs, ontology fragment, and corresponding table sche…
Figure 5
Figure 5. Figure 5: (a) Visual representation of the chemotherapy delta ontology structure generated by Llama 3.1 from RIGOR from real-world database, generated using WebVOWL 1.1.7. (b) Semantic alignment between ontology classes and database table column names based on embedding-based si…

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. SINT-Flow: Schema Integration using Large Language Model Workflows

    cs.CL 2026-07 conditional novelty 6.0 of 10

    Five LLM operators arranged into workflows fully automate schema integration, including splitting denormalized multi-entity tables, reaching ≥83% mapping F1 on a new 93-table benchmark.

Reference graph

Works this paper leans on

69 extracted references · 46 canonical work pages · cited by 1 Pith paper

  1. [1]

    arXiv preprint arXiv:2412.00608 (2024)

    Abolhasani, M.S., Pan, R.: Leveraging llm for automated ontology extraction and knowledge graph generation. arXiv preprint arXiv:2412.00608 (2024)

    arXiv 2024

  2. [2]

    In: Hong, J., Park, J.W

    Alharbi, R., Tamma, V., Grasso, F., Payne, T.R.: An experiment in retrofitting competency questions for existing ontologies. In: Hong, J., Park, J.W. (eds.) Pro- ceedingsofthe39thACM/SIGAPPSymposiumonAppliedComputing,SAC2024, Avila, Spain, April 8-12, 2024. pp. 1650–1658. ACM (2024).https://doi.org/10. 1145/3605098.3636053,https://doi.org/10.1145/3605098.3636053

    arXiv 2024

  3. [3]

    Antia, M., Keet, C.M.: Automating the generation of competency questions for ontologies with agocqs. In: Ortiz-Rodríguez, F., Villazón-Terrazas, B., Tiwari, S., Bobed,C.(eds.)KnowledgeGraphsandSemanticWeb-5thIberoamericanConfer- ence and 4th Indo-American Conference, KGSWC 2023, Zaragoza, Spain, Novem- ber 13-15, 2023, Proceedings. Lecture Notes in Comput...

    work page doi:10.1007/978-3-031-47745-4_ 2023

  4. [4]

    W3C recommendation27, 1–11 (2012)

    Arenas, M., Bertails, A., Prud’hommeaux, E., Sequeda, J., et al.: A direct mapping of relational data to rdf. W3C recommendation27, 1–11 (2012)

    work page 2012

  5. [5]

    In: European Semantic Web Symposium

    Astrova, I.: Reverse engineering of relational databases to ontologies. In: European Semantic Web Symposium. pp. 327–341. Springer (2004)

    work page 2004

  6. [6]

    In: International Semantic Web Conference

    Babaei Giglou, H., D’Souza, J., Auer, S.: Llms4ol: Large language models for on- tology learning. In: International Semantic Web Conference. pp. 408–427. Springer (2023)

    work page 2023

  7. [7]

    In: International Semantic Web Conference

    Babaei Giglou, H., D’Souza, J., Auer, S.: Llms4ol: Large language models for on- tology learning. In: International Semantic Web Conference. pp. 408–427. Springer (2023) Retrieval-Augmented Generation of Ontologies from Relational Databases 17

    work page 2023

  8. [8]

    In: Proceedings of the 3rd NLP4KGC International Workshop on Natural Language Processing for Knowledge Graph Creation, colo- cated with Semantics

    Bakker, R.M., Di Scala, D.L., de Boer, M.H.: Ontology learning from text: an analysis on llm performance. In: Proceedings of the 3rd NLP4KGC International Workshop on Natural Language Processing for Knowledge Graph Creation, colo- cated with Semantics. pp. 17–19 (2024)

    work page 2024

Show all 69 references
  1. [9]

    Journal of Big Data 8(1), 25 (2021)

    Ben Mahria, B., Chaker, I., Zahi, A.: A novel approach for learning ontology from relational database: from the construction to the evaluation. Journal of Big Data 8(1), 25 (2021)

    2021

  2. [10]

    In: 2024 IEEE International Conference on Consumer Electronics (ICCE)

    Câmara, V., Mendonca-Neto, R., Silva, A., Cordovil, L.: A large language model approach to sql-to-text generation. In: 2024 IEEE International Conference on Consumer Electronics (ICCE). pp. 1–4. IEEE (2024)

    2024

  3. [11]

    Bioinformatics40(3), btae104 (2024)

    Caufield, J.H., Hegde, H., Emonet, V., Harris, N.L., Joachimiak, M.P., Matent- zoglu, N., Kim, H., Moxon, S., Reese, J.T., Haendel, M.A., et al.: Structured prompt interrogation and recursive extraction of semantics (spires): A method for populating knowledge bases using zero-...

    2024

  4. [12]

    Ciroku, F., de Berardinis, J., Kim, J., Meroño-Peñuela, A., Presutti, V., Sim- perl, E.: Revont: Reverse engineering of competency questions from knowl- edge graphs via language models. J. Web Semant.82, 100822 (2024). https://doi.org/10.1016/J.WEBSEM.2024.100822,https://doi.o...

    2024

  5. [13]

    WWW (2012)

    Consortium, W.W.W., et al.: R2rml: Rdb to rdf mapping language. WWW (2012)

    2012

  6. [14]

    Applied Sciences15(4), 2146 (2025)

    Doumanas, D., Soularidis, A., Spiliotopoulos, D., Vassilakis, C., Kotis, K.: Fine- tuning large language models for ontology engineering: A comparative analysis of gpt-4 and mistral. Applied Sciences15(4), 2146 (2025)

    2025

  7. [15]

    arXiv preprint arXiv:2401.08281 (2024)

    Douze, M., Guzhva, A., Deng, C., Johnson, J., Szilvasy, G., Mazaré, P.E., Lomeli, M., Hosseini, L., Jégou, H.: The faiss library. arXiv preprint arXiv:2401.08281 (2024)

    2024 arXiv

  8. [16]

    In: Eu- ropean Semantic Web Conference

    Fathallah, N., Das, A., Giorgis, S.D., Poltronieri, A., Haase, P., Kovriguina, L.: Neon-gpt: a large language model-powered pipeline for ontology learning. In: Eu- ropean Semantic Web Conference. pp. 36–50. Springer (2024)

    2024

  9. [17]

    In: The Twelfth International Conference on Learning Representations (2024)

    Galkin, M., Yuan, X., Mostafa, H., Tang, J., Zhu, Z.: Towards foundation mod- els for knowledge graph reasoning. In: The Twelfth International Conference on Learning Representations (2024)

    2024

  10. [18]

    Information15(2), 68 (2024)

    Galli, C., Donos, N., Calciolari, E.: Performance of 4 pre-trained sentence trans- former models in the semantic query of a systematic review dataset on peri- implantitis. Information15(2), 68 (2024)

    2024

  11. [19]

    arXiv preprint arXiv:2312.10997 (2023)

    Gao, Y., Xiong, Y., Gao, X., Jia, K., Pan, J., Bi, Y., Dai, Y., Sun, J., Wang, M., Wang, H.: Retrieval-augmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997 (2023)

    2023 arXiv

  12. [20]

    arXiv preprint arXiv:2407.21783 (2024)

    Grattafiori, A., Dubey, A., Jauhri, A., Pandey, A., Kadian, A., Al-Dahle, A., Let- man, A., Mathur, A., Schelten, A., Vaughan, A., et al.: The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

    2024 arXiv

  13. [21]

    In: Proc

    Gruninger, M.: Methodology for the design and evaluation of ontologies. In: Proc. IJCAI’95, Workshop on Basic Ontological Issues in Knowledge Sharing (1995)

    1995

  14. [22]

    CoRRabs/2411.15594 (2024).https://doi.org/10.48550/ARXIV.2411.15594,https://doi.org/10

    Gu, J., Jiang, X., Shi, Z., Tan, H., Zhai, X., Xu, C., Li, W., Shen, Y., Ma, S., Liu, H., Wang, Y., Guo, J.: A survey on llm-as-a-judge. CoRRabs/2411.15594 (2024).https://doi.org/10.48550/ARXIV.2411.15594,https://doi.org/10. 48550/arXiv.2411.15594

    2024 doi arXiv

  15. [23]

    arXiv preprint arXiv:2411.15594 (2024) 18 M

    Gu, J., Jiang, X., Shi, Z., Tan, H., Zhai, X., Xu, C., Li, W., Shen, Y., Ma, S., Liu, H., et al.: A survey on llm-as-a-judge. arXiv preprint arXiv:2411.15594 (2024) 18 M. Nayyeri et al

    2024 arXiv

  16. [24]

    arXiv preprint arXiv:2501.12948 (2025)

    Guo, D., Yang, D., Zhang, H., Song, J., Zhang, R., Xu, R., Zhu, Q., Ma, S., Wang, P., Bi, X., et al.: Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948 (2025)

    2025 arXiv

  17. [25]

    Horridge, M., Patel-Schneider, P.F.: Owl 2 web ontology language manchester syn- tax.https://www.w3.org/TR/owl2-manchester-syntax/(2012)

    2012

  18. [26]

    In: The Semantic Web-ISWC 2015: 14th International Semantic Web Conference, Bethlehem, PA, USA, October 11-15, 2015, Proceedings, Part II 14

    Jiménez-Ruiz, E., Kharlamov, E., Zheleznyakov, D., Horrocks, I., Pinkel, C., Skjæveland, M.G., Thorstensen, E., Mora, J.: Bootox: Practical mapping of rdbs to owl 2. In: The Semantic Web-ISWC 2015: 14th International Semantic Web Conference, Bethlehem, PA, USA, October 11-15, ...

    2015

  19. [27]

    In: Garoufallou, E., Fallucchi, F., Luca, E.W.D

    Keet, C.M., Mahlaza, Z., Antia, M.: Claro: A controlled language for author- ing competency questions. In: Garoufallou, E., Fallucchi, F., Luca, E.W.D. (eds.) Metadata and Semantic Research - 13th International Conference, MTSR 2019, Rome, Italy, October 28-31, 2019, Revised S...

    2019 doi

  20. [28]

    In: Extended semantic web conference

    Knoblock, C.A., Szekely, P., Ambite, J.L., Goel, A., Gupta, S., Lerman, K., Muslea, M., Taheriyan, M., Mallick, P.: Semi-automatically mapping structured sources into the semantic web. In: Extended semantic web conference. pp. 375–

  21. [29]

    In: Proceedings IEEE International Symposium on Bio- Informatics and Biomedical Engineering

    Kohler, J., Lange, M., Hofestadt, R., Schulze-Kremer, S.: Logical and seman- tic database integration. In: Proceedings IEEE International Symposium on Bio- Informatics and Biomedical Engineering. pp. 77–80. IEEE (2000)

    2000

  22. [30]

    Journal Name (2024)

    Kommineni, V.K., König-Ries, B., Samuel, S.: Towards the automation of knowl- edge graph construction using large language models. Journal Name (2024)

    2024

  23. [31]

    Kommineni, V.K., König-Ries, B., Samuel, S.: From human experts to machines: An llm supported approach to ontology and knowledge graph construction (2024), https://arxiv.org/abs/2403.08345

    2024 arXiv

  24. [32]

    Advances in neural information processing systems 33, 9459–9474 (2020)

    Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W.t., Rocktäschel, T., et al.: Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems 33, 9459–9474 (2020)

    2020

  25. [34]

    arXiv preprint arXiv:2503.05388 (2025)

    Lippolis, A.S., Saeedizade, M.J., Keskisärkkä, R., Zuppiroli, S., Ceriani, M., Gangemi, A., Blomqvist, E., Nuzzolese, A.G.: Ontology generation using large language models. arXiv preprint arXiv:2503.05388 (2025)

    2025 arXiv

  26. [35]

    we need structured output

    Liu, M.X., Liu, F., Fiannaca, A.J., Koo, T., Dixon, L., Terry, M., Cai, C.J.: " we need structured output": Towards user-centered constraints on large language model output. In: Extended Abstracts of the CHI Conference on Human Factors in Computing Systems. pp. 1–9 (2024)

    2024

  27. [36]

    Information Processing & Management61(5), 103809 (2024)

    Liu, Y., Li, D., Wang, K., Xiong, Z., Shi, F., Wang, J., Li, B., Hang, B.: Are llms good at structured outputs? a benchmark for evaluating structured output capabilities in llms. Information Processing & Management61(5), 103809 (2024)

    2024

  28. [37]

    In: THE WEB CONFERENCE 2025 (2025) Retrieval-Augmented Generation of Ontologies from Relational Databases 19

    Liu, Z., Gan, C., Wang, J., Zhang, Y., Bo, Z., Sun, M., Chen, H., Zhang, W.: Ontotune: Ontology-driven self-training for aligning large language models. In: THE WEB CONFERENCE 2025 (2025) Retrieval-Augmented Generation of Ontologies from Relational Databases 19

    2025

  29. [38]

    In: The Thirty-eighth Annual Conference on Neural Information Processing Systems (2024)

    Lo, A., Jiang, A.Q., Li, W., Jamnik, M.: End-to-end ontology learning with large language models. In: The Thirty-eighth Annual Conference on Neural Information Processing Systems (2024)

    2024

  30. [39]

    Vietnam Journal of Computer Science9(01), 31–57 (2022)

    Ma, C., Molnár, B.: Ontology learning from relational database: Opportunities for semantic information integration. Vietnam Journal of Computer Science9(01), 31–57 (2022)

    2022

  31. [40]

    In: 2023 25th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing (SYNASC)

    Mateiu, P., Groza, A.: Ontology engineering with large language models. In: 2023 25th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing (SYNASC). pp. 226–229. IEEE (2023)

    2023

  32. [41]

    In: International conference on web engineering

    de Medeiros, L.F., Priyatna, F., Corcho, O.: Mirror: Automatic r2rml mapping gen- eration from relational databases. In: International conference on web engineering. pp. 326–343. Springer (2015)

    2015

  33. [42]

    Meersman, R.: Ontologies and databases: More than a fleeting resemblance. In: A. d’Atri and M. Missikoff (eds), Proceedings of the OES/SEO 2001 Rome Workshop, LUISS Publications, Rome (2001) (2001)

    2001

  34. [43]

    arXiv preprint arXiv:2308.16622 (2023)

    Meyer, L.P., Frey, J., Junghanns, K., Brei, F., Bulert, K., Gründer-Fahrer, S., Martin, M.: Developing a scalable benchmark for assessing large language models in knowledge graph engineering. arXiv preprint arXiv:2308.16622 (2023)

    2023 arXiv

  35. [45]

    In: Interna- tional semantic web conference

    Mihindukulasooriya, N., Tiwari, S., Enguix, C.F., Lata, K.: Text2kgbench: A benchmark for ontology-driven knowledge graph generation from text. In: Interna- tional semantic web conference. pp. 247–265. Springer (2023)

    2023

  36. [46]

    In: Proceedings of the 16th international conference on World Wide Web

    Motik, B., Horrocks, I., Sattler, U.: Bridging the gap between owl and relational databases. In: Proceedings of the 16th international conference on World Wide Web. pp. 807–816 (2007)

    2007

  37. [48]

    In: Handbook on Neurosymbolic AI and Knowledge Graphs, pp

    Norouzi, S.S., Barua, A., Christou, A., Gautam, N., Eells, A., Hitzler, P., Shimizu, C.: Ontology population using llms. In: Handbook on Neurosymbolic AI and Knowledge Graphs, pp. 421–438. IOS Press (2025)

    2025

  38. [49]

    Semantic Web3(2), 169–209 (2012)

    Passant, A., Gandon, F., Alani, H., Spanos, D.E., Stavrou, P., Mitrou, N.: Bringing relational databases into the semantic web: A survey. Semantic Web3(2), 169–209 (2012)

    2012

  39. [50]

    Semantic Web 9(1), 25–52 (2017)

    Pinkel, C., Binnig, C., Jiménez-Ruiz, E., Kharlamov, E., May, W., Nikolov, A., Sasa Bastinos, A., Skjæveland, M.G., Solimando, A., Taheriyan, M., et al.: Rodi: Benchmarking relational-to-ontology mapping generation quality. Semantic Web 9(1), 25–52 (2017)

    2017

  40. [51]

    Lecture Notes in Informatics (LNI), Proceedings-Series of the Gesellschaft fur Informatik (GI)265, 145–164 (2017)

    Pinkel, C., Binnig, C., Jiménez-Ruiz, E., Kharlamov, E., Nikolov, A., Schwarte, A., Heupel, C., Kraska, T.: Incmap: A journey towards ontology-based data inte- gration. Lecture Notes in Informatics (LNI), Proceedings-Series of the Gesellschaft fur Informatik (GI)265, 145–164 (2017)

    2017

  41. [52]

    Scientific Data5, 180178 (09 2018).https://doi.org/10.1038/sdata

    Pollard, T., Johnson, A., Raffa, J., Celi, L., Mark, R., Badawi, O.: The eicu collab- orative research database, a freely available multi-center database for critical care research. Scientific Data5, 180178 (09 2018).https://doi.org/10.1038/sdata. 2018.178

    2018 doi

  42. [53]

    International 20 M

    Poveda-Villalón, M., Gómez-Pérez, A., Suárez-Figueroa, M.C.: Oops! (ontol- ogy pitfall scanner!): An on-line tool for ontology evaluation. International 20 M. Nayyeri et al. Journal on Semantic Web and Information Systems (IJSWIS)10(2), 7– 34 (2014).https://doi.org/10.4018/IJS...

    2014 doi

  43. [54]

    arXiv preprint arXiv:2312.00326 (2023)

    Qiang, Z., Wang, W., Taylor, K.: Agent-om: Leveraging llm agents for ontology matching. arXiv preprint arXiv:2312.00326 (2023)

    2023 arXiv

  44. [55]

    (eds.) The Semantic Web: ESWC 2024 Satellite Events - Hersonissos, Crete, Greece, May 26-30, 2024, Proceed- ings, Part I

    Rebboud, Y., Tailhardat, L., Lisena, P., Troncy, R.: Can llms generate competency questions? In: Meroño-Peñuela, A., Corcho, Ó., Groth, P., Simperl, E., Tamma, V., Nuzzolese, A.G., Poveda-Villalón, M., Sabou, M., Presutti, V., Celino, I., Revenko, A., Raad, J., Sartini, B., Li...

    2024

  45. [56]

    Advances in Neural Information Processing Systems37, 21330– 21341 (2024)

    Robinson, J., Ranjan, R., Hu, W., Huang, K., Han, J., Dobles, A., Fey, M., Lenssen, J.E., Yuan, Y., Zhang, Z., et al.: Relbench: A benchmark for deep learning on rela- tional databases. Advances in Neural Information Processing Systems37, 21330– 21341 (2024)

    2024

  46. [57]

    Future Internet15(6), 192 (2023)

    Roumeliotis, K.I., Tselikas, N.D.: Chatgpt and open-ai models: A preliminary re- view. Future Internet15(6), 192 (2023)

    2023

  47. [58]

    arXiv preprint arXiv:2504.00752 (2025)

    Sadruddin, S., D’Souza, J., Poupaki, E., Watkins, A., Giglou, H.B., Rula, A., Kara- sulu, B., Auer, S., Mackus, A., Kessels, E.: Llms4schemadiscovery: A human-in- the-loop workflow for scientific schema mining with large language models. arXiv preprint arXiv:2504.00752 (2025)

    2025 arXiv

  48. [59]

    In: European Semantic Web Conference

    Saeedizade, M.J., Blomqvist, E.: Navigating ontology development with large lan- guage models. In: European Semantic Web Conference. pp. 143–161. Springer (2024)

    2024

  49. [60]

    In: Proceedings of the 21st international conference on World Wide Web

    Sequeda, J.F., Arenas, M., Miranker, D.P.: On directly mapping relational databases to rdf and owl. In: Proceedings of the 21st international conference on World Wide Web. pp. 649–658 (2012)

    2012

  50. [61]

    arXiv preprint arXiv:2412.15235 (2024)

    Sharma, K., Kumar, P., Li, Y.: Og-rag: Ontology-grounded retrieval-augmented generation for large language models. arXiv preprint arXiv:2412.15235 (2024)

    2024 arXiv

  51. [62]

    Journal of Web Semantics p

    Shimizu, C., Hitzler, P.: Accelerating knowledge graph and ontology engineering with large language models. Journal of Web Semantics p. 100862 (2025)

    2025

  52. [63]

    Information Processing & Man- agement62(3), 104042 (2025)

    Val-Calvo, M., Aranguren, M.E., Mulero-Hernández, J., Almagro-Hernández, G., Deshmukh, P., Bernabé-Díaz, J.A., Espinoza-Arias, P., Sánchez-Fernández, J.L., Mueller, J., Fernández-Breis, J.T.: Ontogenix: Leveraging large language models for enhanced ontology engineering from da...

    2025

  53. [64]

    Vergoulis, A., et al.: Data governance in the era of the web of data: generate, man- age, preserve, share and protect resources in the web of data. Tech. rep., Technical Report (2014)

    2014

  54. [65]

    Volz, R., Handschuh, S., Staab, S., Stojanovic, L., Stojanovic, N.: Unveiling the hidden bride: deep annotation for mapping and migrating legacy data to the se- mantic web. J. Web Semant.1(2), 187–206 (2004).https://doi.org/10.1016/J. WEBSEM.2003.11.005,https://doi.org/10.1016...

    2004 doi

  55. [66]

    http://www.w3.org/TR/owl2-profiles/(2012), w3C Recommendation

    W3C OWL Working Group: Owl 2 web ontology language profiles (second edition). http://www.w3.org/TR/owl2-profiles/(2012), w3C Recommendation

    2012

  56. [67]

    In: Ois@ ijcai (2001) Retrieval-Augmented Generation of Ontologies from Relational Databases 21

    Wache, H., Voegele, T., Visser, U., Stuckenschmidt, H., Schuster, G., Neumann, H., Hübner, S.: Ontology-based integration of information-a survey of existing ap- proaches. In: Ois@ ijcai (2001) Retrieval-Augmented Generation of Ontologies from Relational Databases 21

    2001

  57. [68]

    In: Koyejo, S., Mohamed, S., Agarwal, A., Bel- grave, D., Cho, K., Oh, A

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E.H., Le, Q.V., Zhou, D.: Chain-of-thought prompting elicits reasoning in large language models. In: Koyejo, S., Mohamed, S., Agarwal, A., Bel- grave, D., Cho, K., Oh, A. (eds.) Advances in Neural Informat...

    2022

  58. [69]

    IEEE transactions on neural networks and learning systems32(1), 4–24 (2020)

    Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., Yu, P.S.: A comprehensive survey on graph neural networks. IEEE transactions on neural networks and learning systems32(1), 4–24 (2020)

    2020

  59. [70]

    In: The Thirteenth International Conference on Learning Represen- tations (2025),https://openreview.net/forum?id=md9qolJwLl

    Xiong, B., Staab, S.: From tokens to lattices: Emergent lattice structures in lan- guage models. In: The Thirteenth International Conference on Learning Represen- tations (2025),https://openreview.net/forum?id=md9qolJwLl

    2025

  60. [71]

    arXiv preprint arXiv:2303.18223 (2023)

    Zhao, W.X., Zhou, K., Li, J., Tang, T., Wang, X., Hou, Y., Min, Y., Zhang, B., Zhang, J., Dong, Z., et al.: A survey of large language models. arXiv preprint arXiv:2303.18223 (2023)

    2023 arXiv

  61. [72]

    Score: X.X

    Zhou, H., Nova, A., Larochelle, H., Courville, A., Neyshabur, B., Sedghi, H.: Teach- ing algorithmic reasoning via in-context learning. arXiv preprint arXiv:2211.09066 (2022) 22 M. Nayyeri et al. 6 Appendix This appendix provides supplementary material supporting the methods, ...

    2022 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.