Pith. sign in

REVIEW 4 major objections 6 minor 34 references

A schema-first planning server lets language models query curated mathematical-model knowledge in plain language while keeping every answer tied to expert-curated facts.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-31 12:46 UTC pith:673DBL7K

load-bearing objection Solid reusable MCP pattern for Wikibase KGs with open code and two real demos; the headline R@10 number is weaker than the abstract implies and should not be treated as strong evidence. the 4 major comments →

arxiv 2607.24512 v1 pith:673DBL7K submitted 2026-07-27 cs.AI cs.DL

Making Mathematical Knowledge Explainable, Accessible and Interoperable Through Large Language Model Integration

Jan Range , Bj\"orn Schembera , Dominik G\"oddeke This is my paper
classification cs.AI cs.DL
keywords mathematical modelsknowledge graphslarge language modelsModel Context ProtocolSPARQLSteiner treeschema retrievalFAIR
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Mathematical models are how many fields formalize problems, but curated model knowledge bases are hard to use: people must wrestle with web UIs or write SPARQL and API calls, and linking those bases to real research datasets is still awkward. This paper claims that wrapping such a knowledge graph in a Model Context Protocol server fixes that. The server first retrieves the relevant schema fragments with dual vector encoders, plans the cheapest valid joins with a Steiner tree, and only then lets a language model write SPARQL against a blueprint of legal paths. Instantiated on a Wikibase-based mathematical model graph, the pattern is meant to transfer to other graphs of the same kind. Schema recall averages 0.89 on typical researcher questions, and two end-to-end cases show an agent matching an enzyme-kinetics dataset to the right rate-law family and pulling coupling conditions plus linked biographical facts for a continuum-mechanics model from ordinary prompts.

Core claim

Integrating large language models with a curated mathematical-model knowledge graph through an MCP server that exposes vector-indexed schema retrieval and a Steiner-tree join planner enables epistemically grounded natural-language access, improves explainability and navigability beyond the standard interface, and simplifies interoperability with external research-data tools—without rewriting the underlying graph.

What carries the argument

The Explore_Ontology pipeline: dual-encoder retrieval (dense embeddings plus ColBERT re-ranking) over typed schema collections, query-conditioned edge costs, and a Steiner-tree join planner that returns a TOON-encoded blueprint the agent is expected to follow before any SPARQL runs.

Load-bearing premise

The language model will actually obey the Steiner join blueprint when it writes SPARQL, even though the protocol is stateless and cannot force later steps to match the earlier plan.

What would settle it

Build a controlled natural-language QA set with gold SPARQL and answers over the model graph; compare end-to-end accuracy of the full pipeline against a plain LLM-plus-SPARQL baseline and a no-Steiner ablation. The central claim fails if answer correctness does not clearly beat those baselines once blueprint adherence is optional.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Users can ask about models, equations, coupling conditions, and linked provenance in ordinary language without learning SPARQL or federation syntax.
  • Agents can join research datasets in external repositories to formal model descriptions through multi-server natural-language workflows.
  • The same schema-first wrapper can be reused on other Wikibase-style knowledge graphs without changing the stored data.
  • Model-based research outputs become easier to explain because answers rest on curated epistemic metadata rather than unconstrained generated text.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A thin validation step that re-checks generated SPARQL against the prior Steiner blueprint would close the reliability gap left by stateless MCP.
  • Formulation-and-equation classes, where schema recall was weakest, are the highest-yield target for better retrieval or richer entity text.
  • The reverse path—extracting models from papers and writing them into the graph—would complete a closed curation loop the paper already flags as next work.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper presents an MCP (Model Context Protocol) server that wraps the MathModDB Wikibase knowledge graph so that LLM agents can access it via natural language. The architecture combines (i) an offline dual-encoder (dense + ColBERT re-ranking) vector index over four typed schema collections (classes, object properties, data properties, qualifier properties), (ii) a runtime query-planning pipeline that conditions a schema graph with query-dependent edge costs (relevance, frequency, hub penalties) and computes a minimum-weight Steiner tree as a join blueprint, and (iii) an MCP interface exposing Explore_Ontology plus guarded SPARQL execution tools. Evaluation consists of schema-retrieval recall (mean R@10 = 0.89 over 11 hand-curated queries in four topical groups) and two qualitative use cases: linking a Dataverse enzyme-kinetics dataset to a kinetic model in MathModDB, and retrieving coupling conditions and biographical data for the Stokes–Darcy model, including implicit federation to Wikidata. The authors claim improved accessibility, explainability, and interoperability over the standard Wikibase interface, with the architecture transferable to any Wikibase instance.

Significance. If the results hold, this is a useful and timely contribution to the Wikibase/LOD ecosystem: a concrete, open-source integration pattern (code at github.com/MaRDI4NFDI/MathModDB-MCP) for grounding LLM agents in curated knowledge graphs, directly relevant to the workshop's community. The paper ships unusually good reproducibility artifacts for a systems demo: public source code, an evaluation pipeline, shared LLM chat transcripts for both use cases, and the reference SPARQL queries. The Steiner-tree join blueprint as a generation-time prior for SPARQL synthesis is a clean, well-motivated idea, and the architecture and method sections are specified clearly enough to reimplement. The two use cases — Dataverse↔MathModDB interoperability and federated biographical lookup via plain prompts — are convincing demonstrations of the accessibility claim. The main weakness is that the single quantitative result does not currently support the reliability claim it is asked to carry, and end-to-end query correctness is demonstrated only anecdotally. These are fixable within the manuscript's scope.

major comments (4)
  1. [§5, Table 1] The paper's only quantitative result is compromised by a ceiling effect that the manuscript does not address. Section 2.1 states MathModDB has exactly 8 classes. Section 5 defines R@10 per category as the fraction of expected schema elements appearing in the top-10 list of the corresponding typed Qdrant collection (Section 3.1). For the class collection, a top-10 list over an 8-item index contains every class by construction, so class-level R@10 is 1.0 regardless of retrieval quality. The sizes of the object/data/qualifier property collections are never reported; for an 8-class, 26k-statement ontology these are plausibly also small enough that top-10 lists saturate. As a concrete test the authors should (a) report the cardinality of each of the four collections, (b) report R@k at k well below collection size (or switch to a metric such as MRR/nDCG on the ranked list), and (c) break the 0
  2. [§5] The evaluation provides no baseline or ablation, so the contribution of the paper's two main technical components is unmeasured. The 11 queries and their expected schema elements are hand-curated by the system builders and were 'chosen to exercise different parts of the ontology' — i.e., written against the same schema the index was built from. Minimal additions that would make the claim interpretable: a dense-only retriever (no ColBERT re-ranking), a lexical baseline (BM25 over the same retrieval strings), and a no-conditioning ablation (Steiner on the unweighted graph, f_sim = f_freq = f_hub = 1). As things stand, the 0.80 score for the formulations & equations group is the only evidence the metric discriminates at all, and there is no way to tell whether the dual-encoder stack or the edge-cost conditioning in Eqs. (4)–(7) earns anything over trivial retrieval.
  3. [§3.3 vs §6] There is an internal tension about whether the schema-first pipeline is actually enforced. Section 3.3 states Explore_Ontology 'must be invoked prior to any SPARQL execution,' but Section 6 concedes that 'since MCP is stateless, the planning and execution calls are independent ... [the validation pass] cannot verify full adherence to the Steiner blueprint computed earlier.' This matters because the central reliability claim is that the Steiner blueprint 'constrains subsequent SPARQL formulation to schema-valid paths' (§4.4). If adherence is advisory, the demonstrated end-to-end correctness rests entirely on two qualitative chat transcripts (Case 1, Case 2) with no controlled QA benchmark (e.g., execution accuracy over a fixed set of gold SPARQL/answer pairs, with and without the blueprint). At minimum the manuscript should quantify, over the evaluation queries, how often the agent's gene
  4. [§4.3, Eqs. (4)–(7)] The query-conditioning scheme of Eqs. (4)–(7) introduces at least six free hyperparameters (α_s, α_p, α_o, γ, ε, α_f, α_h, plus the stop list and top-k), none of whose values are reported in the text, and no sensitivity analysis is given. Since the cost landscape is the mechanism by which retrieval scores shape the join plan, the paper should state the values used in the evaluation and give at least a coarse indication of robustness (e.g., R@10 variation as α_p dominance or γ is perturbed). The code release mitigates this, but the manuscript itself should be self-contained on a load-bearing design choice.
minor comments (6)
  1. [§4.4] The Steiner tree problem is NP-hard; Section 4.4 should state which exact or approximation algorithm is used and give the typical graph size / runtime. The schema graph is presumably tiny (8 classes), so this is not a scalability concern, but one sentence would close the gap.
  2. [§4.4] The restoration of edge direction after computing the Steiner tree on the undirected projection ('best-matching directed counterpart ... using property identity and edge weight as tiebreakers') is underspecified: what happens when both directions exist, or when the undirected edge corresponds to multiple directed edges with different properties? A brief formal statement or example would help.
  3. [§5, Table 1] Table 1 reports a mean input cost of ~36.8k tokens per query, which is substantial; some discussion of cost drivers (TOON payload size vs. few-shot prompting) and of whether the token budget scales with ontology size would strengthen the transferability claim to larger Wikibase instances such as Wikidata.
  4. [§3.3] The TOON serialization format is used as a key interface artifact (§3.3, Fig. 2) but is never explained beyond a GitHub link; a one-sentence description (and why it was chosen over JSON) would help readers.
  5. [various] Typos and small errors: 'similiar' (§1), 'data respository' (§1 and §5), 'posses' (§1), 'PlasmaPphysics' (ref. [16]), '#T okens' in Table 1 header, inconsistent spacing around em-dashes.
  6. [§6, Supplemental Material] The supplementary links mix a named GitHub repository (MathModDB-MCP) with anonymized anonymous.4open.science links for the evaluation pipeline and SPARQL queries; since authorship is declared, the anonymized links should be de-anonymized or the inconsistency explained.

Circularity Check

0 steps flagged

No significant circularity: systems/architecture paper with independent evaluation targets; self-citations are substrate (MathModDB), not load-bearing proofs of the MCP claims.

full rationale

This is an engineering paper proposing an MCP server (vector schema index + Steiner-tree join planner + guarded SPARQL) over MathModDB. It does not claim first-principles physical/mathematical predictions, fitted constants renamed as forecasts, or uniqueness theorems that force the result. The quantitative claim is schema R@10 on 11 hand-curated NL queries plus two qualitative end-to-end demos (Dataverse↔MathModDB enzyme kinetics; Stokes–Darcy coupling/biography). Hand-curated gold labels and author-chosen queries are ordinary (if weak) benchmark practice; they do not make R@10 equal the Steiner objective or the retrieval scores by construction. Self-citations point to prior MathModDB/MaRDI ontology and KG work that is the wrapped substrate, not an unverified uniqueness or ansatz imported to forbid alternatives. The paper’s own caveat that MCP is stateless so execution cannot fully enforce the earlier Steiner blueprint is an honesty limitation, not a circular reduction. Ceiling-effect concerns about R@10 given only eight classes are evaluation-validity issues, not circularity. No step reduces a claimed derivation to its inputs by definition or fit.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 2 invented entities

Load-bearing content is engineering assumptions and retrieval/planning hyperparameters, not physical postulates. The central claim rests on standard graph algorithms, embedding models, the MCP tool contract, and the premise that curated MathModDB paths plus schema-first prompting yield trustworthy answers. No new particles or forces; free parameters are IR/planning knobs and evaluation design choices.

free parameters (4)
  • Retrieval top-k and dual-encoder stack (text-embedding-3-large + ColBERT v2) = R@10 evaluated at 10; encoders fixed as stated
    Candidate pool size and encoder choice directly determine what enters the Steiner terminals; not derived, selected for recall/precision tradeoff.
  • Edge-cost mixture weights α_s, α_p, α_o with α_p dominant; sharpening γ; ε floor
    Query-conditioned schema costs in Eq. (4)–(5) depend on these hand-set factors; they shape which joins the Steiner tree prefers.
  • Frequency and hub penalties α_f, α_h and stop-list of generic properties
    ffreq and fhub plus exclusion of high-frequency properties (e.g., instance of) are curated knobs that prevent shortcut paths; not learned from a held-out QA objective in the paper.
  • SPARQL result limit of 100 and hand-curated expected schema sets for 11 queries = 11 queries; limit 100
    Hard cap and gold schema lists define the reported R@10; different gold or limits would change the headline metric.
axioms (5)
  • domain assumption Wikibase RDF projection can be treated as a directed schema graph of classes, object/data/qualifier properties for planning.
    Section 4.2 adopts the endpoint’s RDF view so Steiner planning applies; correctness of plans assumes this projection matches executable SPARQL.
  • ad hoc to paper Minimum-weight Steiner tree on the undirected conditioned schema is a good join blueprint for subsequent LLM SPARQL generation.
    Section 4.4 equates the planner output with reduced combinatorial search; not proved optimal for NL→SPARQL accuracy, motivated as classical Steiner structure.
  • domain assumption Schema-first MCP tool use (Explore_Ontology before SPARQL) plus curated MathModDB content yields epistemically safer answers than unconstrained LLM generation.
    Core framing in §§1, 3, 6; safety is argued from grounding in expert-curated paths rather than measured hallucination rates.
  • standard math Standard Steiner tree in graphs and dual-encoder retrieval (dense cosine + ColBERT MaxSim) behave as in the cited IR/graph literature.
    Methods cite Hwang & Richards (Steiner) and Khattab & Zaharia (ColBERT); used as off-the-shelf machinery.
  • domain assumption External Linked Open Data links already present in MathModDB (e.g., to Wikidata) suffice for biographical federation without user-written SERVICE clauses.
    Case 2 relies on following pre-existing external links rather than general arbitrary federation discovery.
invented entities (2)
  • MathModDB MCP server (Explore_Ontology + Steiner join blueprint + TOON subschema tools) independent evidence
    purpose: Mediate NL agents and the MathModDB SPARQL endpoint with schema-constrained query planning.
    Primary artifact introduced by the paper; existence is evidenced by the public repo and demos, but ‘epistemic safety’ is not an independently measured physical entity.
  • Query-conditioned schema cost landscape (sim × freq × hub reweighting of ontology edges) no independent evidence
    purpose: Turn static Wikibase schema into a per-query weighted graph for Steiner planning without extra model calls at plan time.
    Constructed representation internal to the planner; validated only indirectly via R@10 and demos, not by external theory.

pith-pipeline@v1.2.0-grok45-kimik3 · 17509 in / 3971 out tokens · 86071 ms · 2026-07-31T12:46:20.505589+00:00 · methodology

0 comments
read the original abstract

Mathematical models are central to formalizing research problems, yet their documentation often falls short of FAIR principles. Knowledge bases such as the Mathematical Model Database (MathModDB) address this gap by providing curated, semantically rich representations of mathematical models. Built on Wikibase, the same open-source infrastructure underlying Wikidata, MathModDB utilizes Semantic Web technologies to support Linked Open Data, collaborative editing, and the storage of semantically enriched metadata, making it a domain-specific knowledge graph within the broader Wikidata ecosystem. However, access to MathModDB currently requires either navigating a complex web interface or proficiency in SPARQL and Wikibase APIs, posing significant barriers for potential users. In addition, the combination of such curated knowledge bases with actual research data stored, e.g., in Dataverse repository instances, remains a challenge. To overcome these limitations, we propose integrating Large Language Models (LLMs) with MathModDB via a Model Context Protocol (MCP) server that exposes a vector-indexed schema retrieval and Steiner-tree-based join planner, combining dialogue-based natural language interaction with curated, epistemically grounded knowledge. Although instantiated on MathModDB, the architecture can be applied to other Wikibase-based systems. We demonstrate that this approach enables epistemically grounded LLM usage, improves model explainability and accessibility beyond what the standard Wikibase interface offers, and simplifies interoperability with external databases and tools, such as Dataverse data repositories. We illustrate the benefits of combining the accessibility of an LLM with the epistemic safety of a curated knowledge base through the adaptability of the MCP protocol by two use cases involving mathematical models in the fields of continuum mechanics and enzyme kinetics.

Figures

Figures reproduced from arXiv: 2607.24512 by Bj\"orn Schembera, Dominik G\"oddeke, Jan Range.

Figure 1
Figure 1. Figure 1: Architecture of the MathModDB MCP server. The LLM agent interacts with three tools [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Query-planning pipeline executed by Explore_Ontology. A natural-language query q is retrieved against the four typed Qdrant collections via a dense encoder fd followed by ColBERT re-ranking (Section 4.3). The normalized scores condition the schema graph through relevance, frequency, and hub penalties. Steiner tree (Section 4.4) selects the minimum-weight subtree S spanning all retrieved terminals, and the … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

34 extracted references · 12 canonical work pages

  1. [1]

    Making mathemati- cal research data FAIR: pathways to improved data sharing

    Conrad TO, Ferrer E, Mietchen D, Pusch L, Stegmüller J, Schubotz M. Making mathemati- cal research data FAIR: pathways to improved data sharing. Scientific Data. 2024;11(1):676. Available from:https://doi.org/10.1038/s41597-024-03480-0

    work page doi:10.1038/s41597-024-03480-0 2024

  2. [2]

    Mathematical research data–an analysis through zbMATH references

    Hulek K, Müller F, Schubotz M, Teschke O. Mathematical research data–an analysis through zbMATH references. European Mathematical Society Magazine. 2019;(113):54-7. Available from:https://doi.org/10.4171/news/113/14

    work page doi:10.4171/news/113/14 2019

  3. [3]

    Bravo MaRDI: a wikibase powered knowledge graph on mathematics

    Schubotz M, Ferrer E, Stegmüller J, Mietchen D, Teschke O, Pusch L, et al. Bravo MaRDI: a wikibase powered knowledge graph on mathematics. In: Proc. ISWC 2023; 2023. Available from:https://wikidataworkshop.github.io/2023/papers/15__novel_bravo_mardi_a_w ikibase_%5B1%5D.pdf. 14

    2023

  4. [4]

    The FAIR Guiding Principles for scientific data management and stewardship

    Wilkinson MD, Dumontier M, Aalbersberg IJ, Appleton G, Axton M, Baak A, et al. The FAIR Guiding Principles for scientific data management and stewardship. Scientific data. 2016;3(1):1-9. Available from:http://doi.org/10.1038/sdata.2016.18

    doi:10.1038/sdata.2016.18 2016

  5. [5]

    Towards a Knowledge Graph for Models and Algorithms in Applied Mathematics

    Schembera B, Wübbeling F, Kleikamp H, Schmidt B, Shehu A, Reidelbach M, et al. Towards a Knowledge Graph for Models and Algorithms in Applied Mathematics. In: Sfakakis M, Garoufallou E, Damigos M, Salaba A, Papatheodorou C, editors. Metadata and Semantic Research. Cham: Springer Nature Switzerland; 2025. p. 95-109

    2025

  6. [6]

    MaRDI: Mathematical Research Data Initiative Proposal

    The MaRDI consortium. MaRDI: Mathematical Research Data Initiative Proposal. Zenodo; 2022

    2022

  7. [7]

    Wikidata: a free collaborative knowledgebase

    Vrandečić D, Krötzsch M. Wikidata: a free collaborative knowledgebase. Communications of the ACM. 2014;57(10):78-85. Available from:https://dx.doi.org/10.1145/2629489

    doi:10.1145/2629489 2014

  8. [8]

    Publishing on the Semantic Web

    Berners-Lee T, Hendler J. Publishing on the Semantic Web. Nature. 2001;410:1023-4

    2001

  9. [9]

    Epis- temic metadata for computational engineering information systems

    Horsch MT, Chiacchiera S, Guevara Carrión G, Kohns M, Müller EA, Šarić D, et al. Epis- temic metadata for computational engineering information systems. In: Formal Ontology in Information Systems: Proceedings of the 13th International Conference (FOIS 2023). SAGE Publications 1 Oliver’s Yard, 55 City Road, London, EC1Y 1SP; 2023. p. 302-17. Available from...

    work page doi:10.3233/faia231136 2023

  10. [10]

    Data, Information, Knowledge, Wisdom, and Explainable Artificial Intelligence

    Schembera B. Data, Information, Knowledge, Wisdom, and Explainable Artificial Intelligence. In: Al Machot F, Horsch MT, Scholze S, editors. Designing the Conceptual Landscape for a XAIR Validation Infrastructure. Cham: Springer Nature Switzerland; 2025. p. 122-32. Avail- able from:https://doi.org/10.1007/978-3-031-89274-5_9

    work page doi:10.1007/978-3-031-89274-5_9 2025

  11. [11]

    MKM: a new interdisciplinary field of research

    Farmer WM. MKM: a new interdisciplinary field of research. SIGSAM Bull. 2004 Jun;38(2):47–52. Available from:https://doi.org/10.1145/1041791.1041795

    arXiv 2004

  12. [12]

    BioModels database: a repository of mathematical models of biological processes

    Chelliah V, Laibe C, Le Novère N. BioModels database: a repository of mathematical models of biological processes. In Silico Systems Biology. 2013:189-99. Available from:http://doi. org/10.1007/978-1-62703-450-0_10

    work page doi:10.1007/978-1-62703-450-0_10 2013

  13. [14]

    Onto MODEL: Ontological mathematical modeling knowledge management

    Suresh P, Joglekar G, Hsu S, Akkisetty P, Hailemariam L, Jain A, et al. Onto MODEL: Ontological mathematical modeling knowledge management. In: Computer Aided Chemical Engineering. vol. 25. Elsevier; 2008. p. 985-90. Available from:http://doi.org/10.1016/S1 570-7946(08)80170-8

    work page doi:10.1016/s1 2008

  14. [15]

    OntoMODEL: ontolog- ical mathematical modeling knowledge management in pharmaceutical product development, 1: conceptual framework

    Suresh P, Hsu SH, Akkisetty P, Reklaitis GV, Venkatasubramanian V. OntoMODEL: ontolog- ical mathematical modeling knowledge management in pharmaceutical product development, 1: conceptual framework. Industrial & Engineering Chemistry Research. 2010;49(17):7758-67. Available from:http://doi.org/10.1021/ie100246w. 15

    work page doi:10.1021/ie100246w 2010

  15. [16]

    Ontological Approach to Formaliza- tion of Knowledge in Computational PlasmaPphysics

    Snytnikov A, Glinskiy B, Zagorulko G, Zagorulko Y. Ontological Approach to Formaliza- tion of Knowledge in Computational PlasmaPphysics. Journal of Physics: Conference Series. 2020;1640:012013. Available from:http://doi.org/10.1088/1742-6596/1640/1/012013

    work page doi:10.1088/1742-6596/1640/1/012013 2020

  16. [17]

    OntoMathEdu: A Linguis- tically Grounded Educational Mathematical Ontology

    Kirillovich A, Nevzorova O, Falileeva M, Lipachev E, Shakirova L. OntoMathEdu: A Linguis- tically Grounded Educational Mathematical Ontology. In: Benzmüller C, Miller B, editors. Intelligent Computer Mathematics. Cham: Springer International Publishing; 2020. p. 157-72. Available from:https://doi.org/10.1007/978-3-030-53518-6_10

    work page doi:10.1007/978-3-030-53518-6_10 2020

  17. [18]

    Prerequisite Relationships of the OntoMathEdu Educational Mathematical Ontology

    Kirillovich A, Falileeva M, Nevzorova O, Lipachev E, Dyupina A, Shakirova L. Prerequisite Relationships of the OntoMathEdu Educational Mathematical Ontology. In: Figueroa-García JC, Díaz-Gutierrez Y, Gaona-García EE, Orjuela-Cañón AD, editors. Applied Computer Sci- ences in Engineering. Cham: Springer International Publishing; 2021. p. 517-24. Available f...

    work page doi:10.1007/978-3-030-86702-7_44 2021

  18. [19]

    Ontolo- gies for Models and Algorithms in Applied Mathematics and Related Disciplines

    Schembera B, Wübbeling F, Kleikamp H, Biedinger C, Fiedler J, Reidelbach M, et al. Ontolo- gies for Models and Algorithms in Applied Mathematics and Related Disciplines. In: Garo- ufallou E, Sartori F, editors. Communications in Computer and Information Science. Cham: Springer Nature Switzerland; 2024. p. 161-8. Available from:http://doi.org/10.1007/97 8-...

    doi:10.1007/97 2024

  19. [20]

    FAIR Representation of Mathematical Research Data: MathModDB and MathAlgoDB as Knowledge Graphs for Mathematical Models and Numerical Algorithms

    Schembera B, Wübbeling F, Shehu A, Biedinger C, Fiedler J, Reidelbach M, et al.. FAIR Representation of Mathematical Research Data: MathModDB and MathAlgoDB as Knowledge Graphs for Mathematical Models and Numerical Algorithms . Zenodo; 2025. Available from: https://doi.org/10.5281/zenodo.16735911

    work page doi:10.5281/zenodo.16735911 2025

  20. [21]

    Mathematics in Wikidata

    Scharpf P, Schubotz M, Gipp B. Mathematics in Wikidata. In: Proc. ISWC 2021; 2021

    2021

  21. [22]

    Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions

    Hou X, Zhao Y, Wang S, Wang H. Model Context Protocol (MCP): Landscape, Security Threats, and Future Research Directions. ACM Trans Softw Eng Methodol. 2026 Feb. Just Accepted. Available from:https://doi.org/10.1145/3796519

    doi:10.1145/3796519 2026

  22. [23]

    A survey on model context protocol: Architecture, state-of-the-art, challenges and future directions

    Ray PP. A survey on model context protocol: Architecture, state-of-the-art, challenges and future directions. Authorea Preprints. 2025. Available from:https://doi.org/10.36227/t echrxiv.174495492.22752319/v1

    arXiv 2025

  23. [24]

    Agentic SPARQL: Evaluating SPARQL- MCP-powered Intelligent Agents on the Federated KGQA Benchmark

    Dobriy D, Bauer F, Azzam A, Banerjee D, Polleres A. Agentic SPARQL: Evaluating SPARQL- MCP-powered Intelligent Agents on the Federated KGQA Benchmark. arXiv preprint arXiv:260306582. 2026

    2026

  24. [25]

    ExperienceswithModelContextPro- tocol Servers for Science and High Performance Computing

    PanH,ChardR,MelloR,GramsC,HeT,BraceA,etal. ExperienceswithModelContextPro- tocol Servers for Science and High Performance Computing. arXiv preprint arXiv:250818489. 2025

    2025

  25. [26]

    Enhancing Dspace with Large Language Models: Designing an Integration Framework Using the Model Context Protocol

    Dasgupta T, Neogi M, Mukhopadhyay P. Enhancing Dspace with Large Language Models: Designing an Integration Framework Using the Model Context Protocol. In: Proceedings of the 4th International Symposium on Knowledge Engineering in Digital Library Design (KEDLD- 2025), 235-245.; 2025. Available from:https://doi.org/10.5281/zenodo.17598560. 16

    work page doi:10.5281/zenodo.17598560 2025

  26. [27]

    Community- based biomedical context to unlock agentic systems

    Kuehl M, Schaub DP, Carli F, Heumos L, Fernández-Zapata C, Kaiser N, et al. Community- based biomedical context to unlock agentic systems. bioRxiv. 2025. Available from:https: //www.biorxiv.org/content/early/2025/07/25/2025.07.21.665729

    2025

  27. [28]

    European standardization efforts from FAIR toward explainable-AI-ready data documentation in materials modelling

    Horsch MT, Schembera B, Preisig HA. European standardization efforts from FAIR toward explainable-AI-ready data documentation in materials modelling. In: 2023 3rd International conference on applied artificial intelligence (ICAPAI). IEEE; 2023. p. 1-6

    2023

  28. [29]

    SPINACH: SPARQL-based in- formation navigation for challenging real-world questions

    Liu S, Semnani S, Triedman H, Xu J, Zhao ID, Lam M. SPINACH: SPARQL-based in- formation navigation for challenging real-world questions. In: Findings of the associa- tion for computational linguistics: EMNLP 2024; 2024. p. 15977-6001. Available from: https://doi.org/10.18653/v1/2024.findings-emnlp.938

    doi:10.18653/v1/2024.findings-emnlp.938 2024

  29. [30]

    Knowledge graphs as context sources for llm-based expla- nations of learning recommendations

    Abu-Rasheed H, Weber C, Fathi M. Knowledge graphs as context sources for llm-based expla- nations of learning recommendations. In: 2024 IEEE Global Engineering Education Conference (EDUCON). IEEE; 2024. p. 1-5

    2024

  30. [31]

    Knowledge Graphs as a source of trust for LLM-powered enterprise question answering

    Sequeda J, Allemang D, Jacob B. Knowledge Graphs as a source of trust for LLM-powered enterprise question answering. Journal of Web Semantics. 2025;85:100858. Available from: https://www.sciencedirect.com/science/article/pii/S1570826824000441

    2025

  31. [32]

    ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT

    Khattab O, Zaharia M. ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT. CoRR. 2020;abs/2004.12832. Available from:https://arxiv. org/abs/2004.12832

    Pith/arXiv arXiv 2020

  32. [33]

    Steiner tree problems

    Hwang FK, Richards DS. Steiner tree problems. Networks. 1992;22(1):55-89. Available from: https://onlinelibrary.wiley.com/doi/abs/10.1002/net.3230220105

    doi:10.1002/net.3230220105 1992

  33. [34]

    Kinetic model development forα- aminoesterhydrolase(AEH)-catalyzedsynthesisofβ-lactamantibiotics

    Lagerman CE, Grover MA, Rousseau RW, Bommarius AS. Kinetic model development forα- aminoesterhydrolase(AEH)-catalyzedsynthesisofβ-lactamantibiotics. ChemicalEngineering Journal. 2021;426:131816

    2021

  34. [35]

    Partitioned Coupling vs

    Schmalfuß J, Riethmüller C, Altenbernd M, Weishaupt K, Göddeke D. Partitioned Coupling vs. Monolithic Block-Preconditioning Approaches for Solving Stokes-Darcy Systems. In: 9th International Conference on Computational Methods for Coupled Problems in Science and Engineering (COUPLED PROBLEMS 2021); 2021

    2021