Pith. sign in

REVIEW 3 major objections 4 minor 65 references

AuthentiCity: A Multi-Source Provenance-Aware Knowledge Graph and Benchmark for 3D City Models

T0 review · 3 major / 4 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read A five-city 3D knowledge graph makes data provenance part of the query, so urban AI can be tested on whether it trusts the right source.

desk verdict A genuinely useful provenance-aware urban KG and benchmark, but the OSM matching threshold is validated by a near-tautological proxy — should be fixed or heavily caveated before release. read the letter →

arxiv 2607.25243 v1 pith:FSTQCMMV submitted 2026-07-28 cs.DB

Huynh Duc An Son Nguyen , Lukas Arzoumanidis , Youness Dehbi This is my paper · ORCID
classification cs.DB
keywords knowledgegraph3Dcitymodelsprovenancebenchmarktext-to-queryrepresentationlearningdatafusionpolygonmatching
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

AuthentiCity is a five-city, 180M-node, provenance-aware 3D city knowledge graph released as a reusable resource, and its central argument is that provenance should be part of the query rather than hidden in documentation. It integrates authoritative cadastral models, crowd-sourced maps, ML-predicted roof materials, and reconstructed 3D geometry under three rules: derived values never overwrite authoritative ones, external objects stay separate nodes, and every fusion edge stores its confidence. On this graph the paper builds two benchmark families — natural-language-to-query and graph representation learning — that test whether a system can tell sources apart, notice coverage gaps, and refuse unanswerable questions. Existing text-to-query and urban benchmarks do not exercise these abilities, so the contribution is a missing testbed for provenance-aware reasoning over urban digital twins.

What carries the argument

The labeled property graph is the load-bearing object. Authoritative features become nodes with coordinates preserved and an R-tree spatial index; crowd-sourced objects stay separate nodes; ENRICHED_BY edges — storing overlap ratio and Jaccard index — record that two objects describe the same place. Connected components of the overlap graph define correspondence classes, and only the primary match propagates source-prefixed properties to the authoritative node. Source, confidence, and coverage are recoverable from node and edge types, so trust is queryable rather than documented.

What would settle it

Take a random sample of roughly 200 accepted and 200 rejected footprint pairs in one city, have an image interpreter label whether each pair is truly the same building, and compare with the tau=0.3 ENRICHED_BY edges; a substantial mismatch rate would show that the geometric threshold encodes a biased notion of cross-source truth. A cheaper check is to verify the sensitivity claim by re-sweeping tau and confirming that the mean primary-match Jaccard really stays constant.

Watch

Extended reading notes

Core claim

The central claim: a multi-source 3D city graph can encode provenance structurally, so trust becomes queryable. Authoritative buildings stay nodes; crowd-sourced objects stay separate nodes; ENRICHED_BY edges carry overlap and Jaccard confidence; derived attributes live under source-prefixed keys. The paper derives two benchmark families: a query family for spatial, cross-source, provenance-filtered, coverage-aware, and infeasible questions, and a learning family comparing provenance-agnostic with provenance-aware embeddings on attribute, classification, and matching tasks. Reported baselines leave large headroom, and the open-weight model never abstains on unanswerable questions.

Load-bearing premise

The whole cross-source layer rests on the geometric rule that two footprints match when their intersection covers at least 30% of the smaller footprint; if that rule pairs buildings that are not the same physical object, every matching link, agreement statistic, and benchmark label built on it is biased.

Editorial extensions

If this is right

  • A text-to-query system can be evaluated on whether it picks the authoritative value over a plausible crowd-sourced or predicted one, and whether it refuses to answer when the requested fact is absent.
  • Graph encoders can be compared with and without source labels and confidence weights on real city-scale data, making the value of provenance awareness measurable rather than assumed.
  • The released overlap graph becomes a resource for polygon-matching research, since many-to-many footprint matching is NP-hard and city-scale ground truth is scarce.
  • The same question can be coverage-aware in one city and infeasible in another, testing whether models adapt to city-specific data availability.
  • The cross-source agreement statistics document how far authoritative and crowd-sourced city data diverge in height, storeys, and roof material, directly useful for data-quality studies.

Reading between the lines

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

  • Implicit in the T3 result but not claimed: learned matching models should be judged on the ambiguous n:m subset, where geometry alone cannot decide and cross-source attributes carry the signal; the paper poses this as an open challenge.
  • The paper leaves the leave-one-city-out ablation of confidence weighting to future work; if that ablation shows confidence weighting matters under distribution shift, it would strengthen the case for provenance-aware encoders in cross-city transfer.
  • A natural next benchmark family would test versioning and update awareness: since every fact carries a source timestamp, models could be asked whether a newer authoritative value supersedes an older one.
  • The multilingual Tokyo layer suggests a clean experiment: ask the same questions in Japanese and English and separate schema-bridging failures from language failures.
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 / 4 minor

Summary. AuthentiCity is a five-city, provenance-aware 3D city knowledge graph and benchmark. The paper describes: (1) construction of labeled property graphs from CityGML for Hamburg, Helsinki, Zurich, New York, and Tokyo; (2) OSM fusion via overlap-thresholded enrichment edges with source-prefixed attributes; (3) ML roof-material and reconstructed LoD3 layers for Hamburg; (4) a 1,394-question text-to-Cypher benchmark spanning spatial, cross-source, provenance-filtered, coverage-aware, infeasible, and multilingual queries; and (5) a representation-learning benchmark (T1/T2/T3) comparing provenance-agnostic and provenance-aware encoders. The central claim is that the dataset makes provenance, source confidence, and coverage first-class, enabling benchmark capabilities absent from prior text-to-query and urban benchmarks. The paper also releases the artifact, code, gold queries, evaluation harness, and datasheet under an archival DOI.

Significance. If the construction pipeline is sound, AuthentiCity is a substantial and reusable resource: 3.6M buildings across heterogeneous CRS, languages, and source vintages, with explicit provenance and coverage, is exactly the kind of artifact the community needs. The paper is unusually transparent: it reports machine-checked verification gates (G1–G5), deterministic gold answers, a held-out split with unpublished queries, 20% human review, paired significance tests, multiple baselines, and explicit limitations. The baseline results are also credible: the in-distribution provenance gains are reported as small, the T3 matching results are negative for learned encoders, and the infeasibility failures are decomposed rather than hidden. The main weakness is the validation of the CityGML–OSM correspondence threshold, which is load-bearing for the cross-source and provenance-aware benchmark claims. That issue is localized and fixable, but it needs to be addressed before the dataset's gold labels can be fully trusted.

major comments (3)
  1. [§3.4 / Appendix C] The justification for τ=0.3 is self-referential. The 'precision proxy' is the mean Jaccard of primary matches, but the primary match is defined as the accepted edge with the largest overlap (Section C). Sweeping τ from 0.1 to 0.7 adds or removes only non-primary edges for buildings whose primary overlap is above the sweep range, so the reported constancy of 0.775 says nothing about whether the correspondences are semantically correct. 'Number of enriched buildings' is likewise not a recall proxy without a ground-truth recall denominator. This is load-bearing because the overlap threshold is the sole criterion for every ENRICHED_BY edge.
  2. [§3.6 and §4–5] Because osm_* attributes are copied to a CityGML building only from the primary match (Section C), any wrong primary match transfers another building's OSM height, storeys, or roof tags onto the cadastral node. This directly contaminates Table 12's agreement/conflict statistics (e.g., Zurich's lower height agreement), the cross-source and coverage-aware gold answers in Section 4, and the T3 ENRICHED_BY labels in Section 5. I am not asking for city-scale ground truth, but the paper should report sensitivity of the released gold answers and Table 12 to τ (e.g., τ ∈ {0.2, 0.5}) and/or provide a stratified human audit of correspondences, especially n:m and low-overlap edges. Without this, the novel cross-source benchmark category encodes an unvalidated matching choice.
  3. [§5.3, Table 7] The cross-city height-transfer table contains unexplained internal comparisons. For Zurich, the leave-one-city-out agnostic/aware R² values (0.581/0.601) are more than double the own-city reference (0.269); for Helsinki the transfer R² is at parity with own-city. This is possible in principle (larger training pool, differing label conventions), but as printed the table invites the reading that cross-city transfer beats within-city training. The paired agnostic-vs-aware comparison is not affected by that confound, but the headline 'improves on four of five held-out cities' needs the own-city column explained or removed.
minor comments (4)
  1. [§5.1] Please clarify whether the provenance-agnostic protocol changes the graph topology (merging source nodes into canonical entities) or only removes source labels and edge weights. The current text says both ('flattens... by merging' and 'sets w=1 and W_s=W'), which makes the claim that the protocols 'differ only in how much provenance structure' ambiguous.
  2. [Table 7 caption] Define 'probe' explicitly and note that it is not comparable to the single-city probe in Table 6; the current parenthetical is easy to miss and the table is otherwise hard to interpret.
  3. [Abstract / §3] The abstract and Section 3 say '180 GiB' for the corpus, while Table 2 lists a Neo4j store size of 176.9 GiB and a compressed dump of 27.2 GiB. Please state which number the abstract refers to.
  4. [Appendix C] Rename 'precision proxy' and 'recall proxy' to 'primary-match stability' and 'enrichment count'; as written they are not precision/recall measurements, which contributes to the confusion in the threshold selection.

Circularity Check

2 steps flagged · score 6.0 of 10

Partial circularity: T3 matching labels and the τ-validation proxy are self-referential; the five-city provenance-aware graph and text-to-query benchmark remain independent.

  1. self definitional [Appendix C (CityGML–OSM Knowledge Graph); τ selection for Section 3.4]
    "Since no city-scale ground-truth CityGML–OSM correspondences exist, we sweep τ∈[0.1,0.7] and evaluate the mean Jaccard of primary matches (precision proxy) and the number of enriched buildings (recall proxy). The mean primary-match Jaccard remains constant (0.775), indicating that τ affects only marginal edges. We therefore select τ=0.3, which provides near-maximal enrichment while preserving correspondence quality."

    The precision proxy used to justify τ is defined by the same construction it is meant to validate: the 'primary match' is the largest-overlap pair selected after thresholding. Lowering τ adds low-overlap edges but leaves the primary match per building unchanged, so constancy of the mean primary-match Jaccard follows from the definition of 'primary match' rather than from independent evidence of correspondence correctness. No external ground truth, human audit, or attribute-based validation is supplied, so the choice of τ rests on a self-referential statistic.

  2. self definitional [Section 5.3 (T3 matching link-prediction paragraph)]
    "For T3 matching, every learned encoder falls far short of non-learned baselines, for a structural reason. ENRICHED_BY correspondences are defined by footprint overlap (median Jaccard 0.842), so on the default nearest-neighbor negatives a spatial-distance rule reaches ROC-AUC 0.95–0.996 while the encoders reach at most 0.57."

    T3 is described as 'link prediction of held-out ENRICHED_BY correspondence edges', but those gold edges were generated in Section 3.4 by exactly the geometric overlap criterion r=area(A∩B)/min(area(A),area(B))≥0.3. Predicting them with a spatial-distance rule therefore recovers the construction rule itself, which is why the non-learned baseline achieves near-perfect AUC. The paper acknowledges this 'structural reason' but still presents T3 as a cross-source matching-prediction task; as a benchmark for semantic matching it is self-referential, since the labels encode the same geometric heuristic used to create them.

full rationale

AuthentiCity is primarily a data-construction paper, and most of its derivation chain is not circular. The CityGML and OSM inputs are external, the provenance model is a stated design choice, the text-to-query gold answers are machine-verified through gates G1–G5 and human-reviewed on 20.2% of questions, and the T1/T2 representation-learning targets are not fitted to the features being predicted. Self-citations to pykci [45] and the roof-material model [4] supply implementation infrastructure and input layers; they are not used to forbid alternatives or to define the benchmark results. However, two load-bearing steps are self-referential. First, the τ=0.3 overlap threshold is selected using a 'precision proxy'—the mean primary-match Jaccard—whose constancy is a property of how primary matches are defined, not evidence of correspondence correctness; every ENRICHED_BY edge, copied osm_* attribute, cross-source statistic, and T3 label inherits this unvalidated geometric criterion. Second, T3's gold labels are the very ENRICHED_BY edges produced by that same overlap construction, so a spatial-distance baseline trivially recovers them; the paper explicitly says the correspondences 'are defined by footprint overlap'. These two points make the cross-source matching benchmark and the threshold justification partially circular, while leaving the main five-city provenance-aware dataset and the natural-language-to-query benchmark substantially independent. Score 6 reflects this partial, sub-task-level circularity rather than a collapse of the paper's central data contribution.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The pipeline rests on a trust model, an overlap-based matching criterion, and a graph-absence infeasibility convention; none is independently benchmarked against external ground truth. These are stated assumptions rather than free parameters in a derivation sense, with τ being the one tuned numeric choice.

free parameters (1)
  • Footprint overlap acceptance threshold τ = 0.3
    Swept over [0.1, 0.7] on the same five-city data; selected for near-maximal enrichment while mean primary-match Jaccard stays 0.775. No ground-truth correspondences exist, so the precision proxy is self-referential (Appendix C / Section 3.4).
assumptions (4)
  • domain assumption Authoritative CityGML data are treated as ground truth and derived values must never overwrite them.
    The entire provenance model (Section 3.1) assumes the cadastral layer is the reference for trust, while OSM, ML, and reconstructed layers are augmentations.
  • domain assumption Overlap coefficient r = area(A∩B)/min(area(A),area(B)) with threshold τ defines CityGML–OSM correspondences.
    Section 3.4 and Figure 4 rely on this geometric criterion as the correspondence signal; no external ground-truth matching is used.
  • domain assumption Absence of an attribute in the graph is proof of infeasibility.
    G4 guard queries return count=0 to label questions infeasible (Section F.4); this identifies graph-level absence, not real-world absence.
  • standard math NP-hardness of many-to-many polygon matching justifies retaining the full overlap graph.
    Section 2.3 cites [42,43] for the complexity result; the design choice follows from that result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AuthentiCity: A Multi-Source Provenance-Aware Knowledge Graph and Benchmark for 3D City Models." pith.science (2026). https://pith.science/paper/FSTQCMMV

@misc{pith2026260725243,
  author       = {Pith},
  title        = {Pith review of: AuthentiCity: A Multi-Source Provenance-Aware Knowledge Graph and Benchmark for 3D City Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FSTQCMMV}},
  note         = {Machine review of arXiv:2607.25243}
}
read the original abstract

Urban digital twins increasingly combine authoritative, crowd-sourced, machine-learned, and reconstructed data with differing reliability, coverage, and semantics. Yet few urban datasets provide a unified representation supporting multi-source integration, provenance tracking, spatial reasoning, and machine learning. We present AuthentiCity, a multi-source, provenance-aware 3D city knowledge graph spanning five cities across three continents (Hamburg, Helsinki, Zurich, New York, and Tokyo) and comprising 180 GiB, 180M nodes, 220M edges, 1.2B properties, and 3.6M buildings. The labeled property graphs integrate authoritative CityGML and OpenStreetMap data for all cities, adding roof-material predictions and reconstructed LoD3 geometry for Hamburg, under a provenance model in which derived information never replaces authoritative data. Confidence-weighted edges resolve cross-source correspondences, constructing canonical urban entities while preserving traceable links to contributing evidence. AuthentiCity is primarily a data contribution. We introduce two benchmark families that demonstrate the tasks enabled by the representation. The first evaluates natural-language-to-query translation beyond conventional text-to-SQL and text-to-Cypher benchmarks, including 3D spatial reasoning, provenance-aware filtering, cross-source agreement and disagreement, coverage-aware aggregation, and infeasible-query detection. The second evaluates graph representation learning through multi-source attribute prediction, node classification, and cross-source matching prediction, enabling comparison of provenance-agnostic and provenance-aware embeddings. Even a strong commercial LLM reaches only 54-69 % execution accuracy and a 7B open-weight model 6-19 %, while the open-weight model never abstains on unanswerable questions.

Figures

Figures reproduced from arXiv: 2607.25243 by the authors.

Figure 1
Figure 1. Hamburg, one of five AuthentiCity datasets: a provenance-aware 3D knowledge graph integrating CityGML, Open￾StreetMap, roof-material predictions (color-coded buildings), and reconstructed LoD3 geometry (orange subgraph). Abstract Urban digital twins increasingly combine authoritative, crowd￾sourced, machine-learned, and reconstructed data with differing reliability, coverage, and semantics. Yet few urban datasets pr… view at source ↗
Figure 2
Figure 2. Overview of AuthentiCity. Four source layers with different trust levels (left) are integrated into a provenance-aware LPG for each city (center). The resulting graphs support two benchmark families with task-specific metrics and dataset-level validation (right). Categories marked (+) are absent from existing text-to-query benchmarks [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Spatial extent of the five AuthentiCity cities. Basemap: © OpenStreetMap contributors. A Datasheet We document AuthentiCity following the datasheets framework of Gebru et al. [22]. Per-city figures referenced below are given in Tables 8 and 10. The cities’ spatial extents are shown in [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: OSM enrichment: CityGML buildings (pink) are linked to OSM buildings (blue) by [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Comparison of the five AuthentiCity graphs. Zurich is the densest, Tokyo the largest, Helsinki the most attribute￾rich, and New York the most OSM-enriched. Geometry ac￾counts for 85–95 % of graph content in every city. G LLM Setup for Text-to-Query Both query-generatio…
Figure 6
Figure 6. Figure 6: 2D t-SNE of Hamburg building embeddings from the T3 matching encoders. Columns show roof type, OSM coverage, [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

65 extracted references · 5 canonical work pages

  1. [1]

    Mahmoud Abdelrahman, Edgardo Macatulad, Binyu Lei, Matias Quintana, Clay- ton Miller, and Filip Biljecki. 2025. What is a Digital Twin anyway? Deriving the definition for the built environment from over 15,000 scientific publications. Building and Environment274 (2025), 112748. doi:10.1016/j.buildenv.2025.112748

    arXiv 2025

  2. [2]

    Jethro Akroyd, Sebastian Mosbach, Amit Bhave, and Markus Kraft. 2021. Uni- versal Digital Twin - A Dynamic Knowledge Graph.Data-Centric Engineering2 (2021), e14. doi:10.1017/dce.2021.10

    doi:10.1017/dce.2021.10 2021

  3. [3]

    2025.Amtliches 3D-Gebäudemodell in der Ausprägung Level of Detail 2 (LoD2-DE)

    Arbeitsgemeinschaft der Vermessungsverwaltungen der Länder der Bundesre- publik Deutschland (AdV). 2025.Amtliches 3D-Gebäudemodell in der Ausprägung Level of Detail 2 (LoD2-DE). https://www.adv-online.de/AdV-Produkte/Weitere- AuthentiCity: A Multi-Source Provenance-Aware Knowledge Graph and Benchmark for 3D City Models Produkte/3D-Gebaeudemodelle-LoD/

    2025

  4. [4]

    Nguyen, Lara Johannsen, Filip Rothaut, Weilian Li, and Youness Dehbi

    Lukas Arzoumanidis, Son H. Nguyen, Lara Johannsen, Filip Rothaut, Weilian Li, and Youness Dehbi. 2025. Object Detection for the Enrichment of Semantic 3D City Models with Roofing Materials.ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information SciencesX-4/W6-2025 (2025), 9–16. doi:10.5194/ isprs-annals-X-4-W6-2025-9-2025

    2025

  5. [5]

    Filip Biljecki, Lawrence Zheng Xiong Chew, Nikola Milojevic-Dupont, and Felix Creutzig. 2021. Open government geospatial data on buildings for planning sustainable and resilient cities.arXiv preprint arXiv:2107.04023(2021). doi:10. 48550/arXiv.2107.04023

    work page Pith review arXiv doi:10.48550/arxiv.2107.04023 2021

  6. [6]

    Filip Biljecki, Yoong Shin Chow, and Kay Lee. 2023. Quality of Crowdsourced Geospatial Building Information: A global Assessment of OpenStreetMap At- tributes.Building and Environment237 (2023), 110295. doi:10.1016/j.buildenv. 2023.110295

    arXiv 2023

  7. [7]

    Filip Biljecki, Hugo Ledoux, Xin Du, Jantien Stoter, Kean Huat Soon, and Victor Khoo. 2016. The most common geometric and semantic errors in CityGML datasets. InISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. IV-2/W1. International Society for Photogrammetry and Remote Sensing (ISPRS), 13–22. doi:10.5194/isprs-ann...

    work page doi:10.5194/isprs-annals-iv-2-w1-13-2016 2016

  8. [8]

    2022.Nutzungsbedingungen für kostenlose Geodaten und Geodienste (OGD) von swisstopo

    Bundesamt für Landestopografie swisstopo. 2022.Nutzungsbedingungen für kostenlose Geodaten und Geodienste (OGD) von swisstopo. https://www.swisstopo. admin.ch/de/nutzungsbedingungen-kostenlose-geodaten-und-geodienste

    2022

Show all 65 references
  1. [9]

    2026.swissBUILDINGS3D 3.0 Beta

    Bundesamt für Landestopografie swisstopo. 2026.swissBUILDINGS3D 3.0 Beta. https://www.swisstopo.admin.ch/de/landschaftmodell-swissbuildings3d- 3-0-beta

    2026

  2. [10]

    2026.Datenlizenz Deutschland – Namensnennung – Version 2.0

    Bundesrepublik Deutschland. 2026.Datenlizenz Deutschland – Namensnennung – Version 2.0. https://www.govdata.de/dl-de/by-2-0

    2026

  3. [11]

    Klau, and Julián Mestre

    Stefan Canzar, Khaled Elbassioni, Gunnar W. Klau, and Julián Mestre. 2011. On Tree-Constrained Matchings and Generalizations. InAutomata, Languages and Programming, Luca Aceto, Monika Henzinger, and Jiří Sgall (Eds.). Springer Berlin Heidelberg, Berlin, Heidelberg, 98–109

    2011

  4. [12]

    Arkadiusz Chadzynski, Nenad Krdzavac, Feroz Farazi, Mei Qi Lim, Shiying Li, Ayda Grisiute, Pieter Herthogs, Aurel von Richthofen, Stephen Cairns, and Markus Kraft. 2021. Semantic 3D City Database - An Enabler for a Dynamic Geospatial Knowledge Graph.Energy and AI6 (2021), 1001...

    2021

  5. [13]

    Shuaichen Chang, Jun Wang, Mingwen Dong, Lin Pan, Henghui Zhu, Alexan- der Hanbo Li, Wuwei Lan, Sheng Zhang, Jiarong Jiang, Joseph Lilien, Steve Ash, William Yang Wang, Zhiguo Wang, Vittorio Castelli, Patrick Ng, and Bing Xi- ang. 2023. Dr.Spider: A Diagnostic Evaluation Bench...

    2023 arXiv

  6. [14]

    Vashu Chauhan, Shobhit Raj, Shashank Mujumdar, Avirup Saha, and Anannay Jain. 2025. Mind the Query: A Benchmark Dataset towards Text2Cypher Task. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track, Saloni Potdar, Lina Rojas...

    2025 doi

  7. [15]

    2026.Creative Commons Attribution 4.0 International License

    Creative Commons. 2026.Creative Commons Attribution 4.0 International License. https://creativecommons.org/licenses/by/4.0

    2026

  8. [16]

    Linfang Ding, Guohui Xiao, Albulen Pano, Mattia Fumagalli, Dongsheng Chen, Yu Feng, Diego Calvanese, Hongchao Fan, and Liqiu Meng. 2025. Integrating 3D City Data through Knowledge Graphs.Geo-spatial Information Science28, 2 (2025), 780–799. arXiv:https://doi.org/10.1080/100950...

    2025

  9. [17]

    Chawla, and Ananthram Swami

    Yuxiao Dong, Nitesh V. Chawla, and Ananthram Swami. 2017. metapath2vec: Scalable Representation Learning for Heterogeneous Networks. InProceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD). 135–144. doi:10.1145/3097983.3098036

    2017

  10. [18]

    Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson

    Vijay Prakash Dwivedi, Chaitanya K. Joshi, Anh Tuan Luu, Thomas Laurent, Yoshua Bengio, and Xavier Bresson. 2023. Benchmarking Graph Neural Networks. Journal of Machine Learning Research24, 43 (2023), 1–48. http://jmlr.org/papers/ v24/22-0567.html

    2023

  11. [19]

    Jie Feng, Tianhui Liu, Yuwei Du, Siqi Guo, Yuming Lin, and Yong Li. 2025. CityGPT: Empowering Urban Spatial Cognition of Large Language Models. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2(Toronto ON, Canada)(KDD ’25). Association ...

    2025

  12. [20]

    Jie Feng, Jun Zhang, Tianhui Liu, Xin Zhang, Tianjian Ouyang, Junbo Yan, Yuwei Du, Siqi Guo, and Yong Li. 2025. CityBench: Evaluating the Capabilities of Large Language Models for Urban Tasks. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mini...

    2025

  13. [21]

    Yanlin Feng, Simone Papicchio, and Sajjadur Rahman. 2025. CypherBench: Towards Precise Retrieval over Full-scale Modern Knowledge Graphs in the LLM Era. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Wanxiang ...

    2025 doi

  14. [22]

    Timnit Gebru, Jamie Morgenstern, Briana Vecchione, Jennifer Wortman Vaughan, Hanna Wallach, Hal Daumé III, and Kate Crawford. 2021. Datasheets for Datasets. Commun. ACM64, 12 (Nov. 2021), 86–92. doi:10.1145/3458723

    2021 doi

  15. [23]

    Aditya Grover and Jure Leskovec. 2016. node2vec: Scalable Feature Learning for Networks. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD). 855–864. doi:10.1145/2939672. 2939754

    2016 doi

  16. [24]

    Kolbe, Claus Nagel, and Karl-Heinz Häfele

    Gerhard Gröger, Thomas H. Kolbe, Claus Nagel, and Karl-Heinz Häfele. 2012. OGC City Geography Markup Language (CityGML) Encoding Standard. Open Geospatial Consortium (OGC). https://portal.ogc.org/files/?artifact_id=47842 OGC 12-019, Version 2.0.0, International Standard

    2012

  17. [25]

    Hamilton, Rex Ying, and Jure Leskovec

    William L. Hamilton, Rex Ying, and Jure Leskovec. 2017. Inductive Representation Learning on Large Graphs. InAdvances in Neural Information Processing Systems 30 (NeurIPS). 1024–1034

    2017

  18. [26]

    2022.3D Models of Helsinki

    Helsingin kaupunginkanslia. 2022.3D Models of Helsinki. https://hri.fi/data/en_ GB/dataset/helsingin-3d-kaupunkimalli

    2022

  19. [27]

    Ziniu Hu, Yuxiao Dong, Kuansan Wang, and Yizhou Sun. 2020. Heterogeneous Graph Transformer. InProceedings of The Web Conference 2020 (WWW). 2704–

    2020

  20. [30]

    Elmehdi Kanna, Jannik Matijevic, Lukas Arzoumanidis, Huynh Duc An Son Nguyen, and Youness Dehbi. 2026. Semantic Enrichment of 3D City Models via Roof Material Classification for Urban Greening and Heat Island Mitigation. SSRN Electronic Journal(2026). doi:10.2139/ssrn.6127041

    2026 doi

  21. [31]

    Bernd Ketzler, Vasilis Naserentin, Fabio Latino, Christopher Zangelidis, Liane Thuvander, and Anders Logg. 2020. Digital Twins for Cities: A State of the Art Review.Built Environment (1978-)46, 4 (2020), 547–573. http://www.jstor.org/ stable/45299343

    2020

  22. [32]

    Kolbe, Tatjana Kutzner, Carl Steven Smyth, Claus Nagel, Carsten Roensdorf, and Charles Heazel

    Thomas H. Kolbe, Tatjana Kutzner, Carl Steven Smyth, Claus Nagel, Carsten Roensdorf, and Charles Heazel. 2021.OGC City Geography Markup Language (CityGML) Part 1: Conceptual Model Standard. Open Geospatial Consortium (OGC). https://www.opengis.net/doc/IS/CityGML-1/3.0 20-010, ...

    2021

  23. [33]

    Daniel Lakens. 2013. Calculating and reporting effect sizes to facilitate cumulative science: A practical primer for t-tests and ANOVAs.Frontiers in Psychology4 (2013), 863. doi:10.3389/fpsyg.2013.00863

    2013

  24. [34]

    2023.Challenges of Urban Digital Twins: A Systematic Review and a Delphi Expert Survey

    Binyu Lei, Patrick Janssen, Jantien Stoter, and Filip Biljecki. 2023.Challenges of Urban Digital Twins: A Systematic Review and a Delphi Expert Survey. Vol. 147. Elsevier BV, 104716. doi:10.1016/j.autcon.2022.104716

    2023

  25. [35]

    Fangyu Lei, Jixuan Chen, Yuxiao Ye, Ruisheng Cao, Dongchan Shin, Hongjin SU, Zhaoqing Suo, Hongcheng Gao, Wenjing Hu, Pengcheng Yin, Victor Zhong, Caim- ing Xiong, Ruoxi Sun, Qian Liu, Sida Wang, and Tao Yu. 2025. Spider 2.0: Evaluat- ing Language Models on Real-World Enterpri...

    2025

  26. [36]

    Jinyang Li, Binyuan Hui, Ge Qu, Jiaxi Yang, Binhua Li, Bowen Li, Bailin Wang, Bowen Qin, Ruiying Geng, Nan Huo, Xuanhe Zhou, Chenhao Ma, Guoliang Li, Kevin Chang, Fei Huang, Reynold Cheng, and Yong- bin Li. 2023. Can LLM Already Serve as a Database Interface? A BIg Bench for L...

    2023

  27. [37]

    Pengyuan Liu and Filip Biljecki. 2022. A review of spatially-explicit GeoAI appli- cations in Urban Geography.International Journal of Applied Earth Observation and Geoinformation112 (2022), 102936. doi:10.1016/j.jag.2022.102936

    2022

  28. [38]

    Liu and C

    S. Liu and C. Wang. 2025. KCitychatBot: A Knowledge Graph Based Chatbot System for Large-scale CityGML Dataset.The International Archives of the Photogrammetry, Remote Sensing and Spatial Information SciencesXLVIII-4/W15- 2025 (2025), 99–105. doi:10.5194/isprs-archives-XLVIII-...

    2025 doi

  29. [39]

    Xinyu Liu, Shuyu Shen, Boyan Li, Nan Tang, and Yuyu Luo. 2025. NL2SQL- BUGs: A Benchmark for Detecting Semantic Errors in NL2SQL Translation. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2(Toronto ON, Canada)(KDD ’25). Association f...

    2025

  30. [40]

    Yu Liu, Jingtao Ding, Yanjie Fu, and Yong Li. 2023. UrbanKG: An Urban Knowl- edge Graph System.ACM Trans. Intell. Syst. Technol.14, 4, Article 60 (May 2023), 25 pages. doi:10.1145/3588577

    2023 doi

  31. [41]

    2025.3D-Gebäudemodell LoD2-DE Hamburg

    Metaver Metadatenverbund, Landesbetrieb Geoinformation und Vermes- sung (LGV) Hamburg. 2025.3D-Gebäudemodell LoD2-DE Hamburg. https: //metaver.de/trefferanzeige?cmd=doShowDocument&docuuid=2C1F2EEC- Nguyen et al. CF9F-4D8B-ACAC-79D8C1334D5E

    2025

  32. [42]

    Alexander Naumann, Annika Bonerath, and Jan-Henrik Haunert. 2024. Many- To-Many Polygon Matching à La Jaccard. In32nd Annual European Symposium on Algorithms (ESA 2024) (Leibniz International Proceedings in Informatics (LIPIcs), Vol. 308), Timothy Chan, Johannes Fischer, John ...

    2024 doi

  33. [43]

    Alexander Naumann, Annika Bonerath, and Jan-Henrik Haunert. 2025. Scalable Many-to-many Building Footprint Matching.Information Fusion124 (2025), 103360. doi:10.1016/j.inffus.2025.103360

    2025

  34. [44]

    2016.3-D Build- ing Model

    New York City Office of Technology and Innovation (OTI). 2016.3-D Build- ing Model. https://github.com/CityOfNewYork/nyc-geo-metadata/blob/main/ Metadata/Metadata_3DBuildingModel.md

    2016

  35. [45]

    Huynh Duc An Son Nguyen, Lukas Arzoumanidis, and Youness Dehbi. 2026. pykci: A Compact Urban Knowledge Graph for Semantic and Spatial Queries using LLMs. arXiv:2607.01605 [cs.DB] https://arxiv.org/abs/2607.01605

    2026 arXiv

  36. [46]

    Yansong Ning and Hao Liu. 2024. UrbanKGent: A Unified Large Language Model Agent Framework for Urban Knowledge Graph Construction. InAdvances in Neural Information Processing Systems, A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang (Eds.), Vol...

    2024 doi

  37. [47]

    Yansong Ning, Hao Liu, Hao Wang, Zhenyu Zeng, and Hui Xiong. 2023. UUKG: Unified Urban Knowledge Graph Dataset for Urban Spatiotemporal Predic- tion. InAdvances in Neural Information Processing Systems, A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine (Eds.)...

    2023

  38. [48]

    2026.NYC Open Data - Overview

    NYC Office of Technology and Innovation (OTI). 2026.NYC Open Data - Overview. https://opendata.cityofnewyork.us/overview/

    2026

  39. [49]

    2026.Open Data Commons Open Database License (ODbL)

    Open Knowledge Foundation. 2026.Open Data Commons Open Database License (ODbL). https://opendatacommons.org/licenses/odbl/

    2026

  40. [50]

    Makbule Gulcin Ozsoy, Leila Messallem, Jon Besga, and Gianandrea Minneci

  41. [51]

    O’Reilly Media, Inc

    Ian Robinson, Jim Webber, and Emil Eifrem. 2015.Graph Databases: New Oppor- tunities for Connected Data. " O’Reilly Media, Inc. "

    2015

  42. [52]

    Kipf, Peter Bloem, Rianne van den Berg, Ivan Titov, and Max Welling

    Michael Schlichtkrull, Thomas N. Kipf, Peter Bloem, Rianne van den Berg, Ivan Titov, and Max Welling. 2018. Modeling Relational Data with Graph Convolutional Networks. InThe Semantic Web – 15th International Confer- ence (ESWC) (Lecture Notes in Computer Science, Vol. 10843). ...

    2018 doi

  43. [53]

    Student. 1908. The Probable Error of a Mean.Biometrika6, 1 (1908), 1–25. doi:10.2307/2331554

    1908 doi

  44. [54]

    Craig Taverner and Andreas Berger. 2025. Neo4j Spatial. https://github.com/ neo4j-contrib/spatial Accessed: June 10, 2026

    2025

  45. [55]

    United Nations Department of Economic and Social Affairs (UN DESA). 2019. World Urbanization Prospects: The 2018 Revision. United Nations. https://www.un- ilibrary.org/content/books/9789210043144

    2019

  46. [56]

    Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. 2018. Graph Attention Networks. InInternational Conference on Learning Representations (ICLR)

    2018

  47. [57]

    Hamilton, Pietro Liò, Yoshua Bengio, and R Devon Hjelm

    Petar Veličković, William Fedus, William L. Hamilton, Pietro Liò, Yoshua Bengio, and R Devon Hjelm. 2019. Deep Graph Infomax. InInternational Conference on Learning Representations (ICLR)

    2019

  48. [58]

    Mohammad Saif Wajid, Hugo Terashima-Marin, Peyman Najafirad, Santiago Enrique Conant Pablos, and Mohd Anas Wajid. 2024. DTwin-TEC: An AI-based TEC District Digital Twin and Emulating Security Events by Leveraging Knowl- edge Graph.Journal of Open Innovation: Technology, Market...

    2024

  49. [59]

    Xiao Wang, Houye Ji, Chuan Shi, Bai Wang, Yanfang Ye, Peng Cui, and Philip S. Yu. 2019. Heterogeneous Graph Attention Network. InThe World Wide Web Conference (WWW). 2022–2032. doi:10.1145/3308558.3313562

    2019

  50. [60]

    Zhu Wang, Fengxia Han, and Shengjie Zhao. 2024. A Survey on Knowledge Graph Related Research in Smart City Domain.ACM Trans. Knowl. Discov. Data 18, 9, Article 223 (Nov. 2024), 31 pages. doi:10.1145/3672615

    2024 doi

  51. [61]

    Frank Wilcoxon. 1945. Individual Comparisons by Ranking Methods.Biometrics Bulletin1, 6 (1945), 80–83. doi:10.2307/3001968

    1945 doi

  52. [62]

    Z. Yao, C. Nagel, M. Kendir, B. Willenborg, and T. H. Kolbe. 2025. The New 3D City Database 5.0 - Advancing 3D City Data Management based on CityGML 3.0.ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences X-4/W6-2025 (2025), 241–248. doi:10.5194...

    2025 doi

  53. [63]

    Zhihang Yao, Claus Nagel, Felix Kunde, György Hudra, Philipp Willkomm, An- dreas Donaubauer, Thomas Adolphi, and Thomas H. Kolbe. 2018.3DCityDB - A 3D Geodatabase Solution for the Management, Analysis, and Visualization of Semantic 3D City Models based on CityGML. Vol. 3. Spri...

    2018 doi

  54. [64]

    Winston Yap, Abraham Noah Wu, Clayton Miller, and Filip Biljecki. 2025. Reveal- ing building operating carbon dynamics for multiple cities.Nature Sustainability 8, 10 (2025), 1199–1210. doi:10.1038/s41893-025-01615-8

    2025 doi

  55. [65]

    Which is the tallest building inside the window⟨wkt⟩?

    Tao Yu, Rui Zhang, Kai Yang, Michihiro Yasunaga, Dongxu Wang, Zifan Li, James Ma, Irene Li, Qingning Yao, Shanelle Roman, Zilin Zhang, and Dragomir Radev. 2018. Spider: A Large-Scale Human-Labeled Dataset for Complex and Cross-Domain Semantic Parsing and Text-to-SQL Task. InPr...

    2018

  56. [2025]

    InPro- ceedings of the Workshop on Generative AI and Knowledge Graphs (GenAIK), Genet Asefa Gesese, Harald Sack, Heiko Paulheim, Albert Merono-Penuela, and Lihu Chen (Eds.)

    Text2Cypher: Bridging Natural Language and Graph Databases. InPro- ceedings of the Workshop on Generative AI and Knowledge Graphs (GenAIK), Genet Asefa Gesese, Harald Sack, Heiko Paulheim, Albert Merono-Penuela, and Lihu Chen (Eds.). International Committee on Computational Li...

    2025

  57. [2710]

    doi:10.1145/3366423.3380027 [28]デジタル庁 . 2024. 公共データ利用規約 (第1.0版). https://www.digital.go.jp/ resources/open_data/public_data_license_v1.0 [29]国土交通省都市局 . 2025.3D都市モデル（Project PLATEAU）ポータルサイ ト. https://www.geospatial.jp/ckan/dataset/plateau

    2024

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.