Pith. sign in

REVIEW 4 major objections 3 minor 72 references

Omni Geometry Representation Learning vs Large Language Models for Geospatial Entity Resolution

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read Omni embeds each geometry type directly and reports up to 12% F1 improvement over point-flattening methods in geospatial entity resolution.

desk verdict The paper's core idea—a uniform geometry encoder for diverse shapes in geospatial entity resolution—is clearly worth a look, but the headline 12% F1 gain is unverifiable from the abstract and rests on a new, undocumented dataset. read the letter →

arxiv 2508.06584 v1 pith:PBZ5RJJX submitted 2025-08-08 cs.DB cs.AI

classification cs.DBcs.AI
keywords geospatialentityresolutiongeometryembeddingmulti-geometrypoints-of-interestlargelanguagemodelsattributeaffinityrecordlinkagespatialdata
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

Geospatial entity resolution is the task of deciding whether two place records refer to the same real-world location. Omni targets a gap in that task: existing neural methods collapse complex geometries such as polygons, polylines, and multipolygons into a single representative point, throwing away shape and extent. The paper's central claim is that a geometry encoder which embeds points, lines, polylines, polygons, and multipolygons into one space, combined with per-attribute text encoding through an Attribute Affinity mechanism, resolves places more accurately. On established point-only benchmarks and on a new diverse-geometry geospatial ER dataset, Omni reports up to a 12% F1 improvement over existing methods. The paper also reports that prompting large language models yields competitive results, suggesting a low-training alternative.

What carries the argument

The omni-geometry encoder is the central object: a neural component that embeds point, line, polyline, polygon, and multi-polygon geometries into a shared vector representation, preserving shape and extent instead of flattening geometries to points. The Attribute Affinity mechanism applies pretrained transformer language models to each textual attribute separately and combines the resulting affinities with the geometry embedding. Together they allow the model to compare the spatial form and descriptive content of place records without manual feature engineering per geometry type.

What would settle it

Re-annotate a random sample of the new diverse-geometry dataset by independent raters and measure agreement; then run an ablation of Omni that replaces the geometry encoder with a single-point or centroid input. If the F1 gap between Omni and the ablation vanishes on the re-annotated sample, or the labels prove to have been generated from the same geometry fields Omni reads, the 12% improvement claim would not survive.

Watch

Extended reading notes

Core claim

The core discovery is a uniform neural representation for heterogeneous geospatial geometries. The omni-geometry encoder converts point, line, polyline, polygon, and multi-polygon geometries into a shared embedding space, preserving each geometry's form rather than reducing it to a centroid. An Attribute Affinity mechanism then combines these geometric embeddings with transformer-based language-model embeddings of individual textual attributes, letting the model compare both spatial form and descriptive content. Tested on point-only datasets and a new diverse-geometry dataset, the model reports up to 12% F1 improvement over existing methods, with the gain attributed to geometry preservation.

Load-bearing premise

The new diverse-geometry dataset is valid: its ground-truth labels are correct, its mix of geometry types reflects real records, and it does not favor Omni simply because the labels were derived from geometry information.

Editorial extensions

If this is right

  • Matching places across databases where one source stores a place as a point and another stores it as a polygon or line becomes feasible without losing the richer geometry.
  • The same approach extends beyond curated point-of-interest sets to any mixed-geometry geospatial dataset, broadening the practical scope of entity resolution.
  • Because textual attributes are modeled individually, records with inconsistent or partially missing text fields can still be compared through the attributes they share.
  • Competitive LLM prompting results imply that zero- or few-shot geospatial entity resolution is a realistic option, lowering the barrier for tasks without labeled training pairs.

Reading between the lines

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

  • If geometry preservation is the active ingredient, then other geometry-aware encoding strategies should show similar gains, making the 12% F1 result a testable property of the principle rather than of Omni's specific architecture.
  • A natural next test is an independent audit of the new diverse-geometry dataset: re-annotation, leakage checks, and comparison against geometry-agnostic baselines would reveal whether the reported improvement transfers to real-world mixed-geometry records.
  • The competitive LLM results suggest that in low-resource settings prompt engineering could substitute for model training; a direct accuracy-versus-cost comparison against Omni would make that trade-off concrete.
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

4 major / 3 minor

Summary. The paper proposes 'Omni', a geospatial Entity Resolution (ER) model that combines an 'omni-geometry encoder' capable of embedding point, line, polyline, polygon, and multi-polygon geometries with a transformer-based language-model component for textual attributes via an 'Attribute Affinity' mechanism. The authors evaluate on existing point-only datasets and a new 'diverse-geometry' dataset they constructed, reporting up to 12% F1 improvement over existing methods. They also explore large language model (LLM) prompting strategies and learning scenarios for geospatial ER, reporting that LLMs are competitive.

Significance. If validated, this work would address a genuine gap: most neural geospatial ER methods flatten complex geometries to points, losing spatial information. A uniform encoder for heterogeneous geometries is a valuable contribution, and the integration with attribute-aware language modeling is sensible. The claim of up to 12% F1 improvement over existing methods, if reproducible, would be significant for the field. The LLM comparison is also timely and practically relevant. However, the paper's central quantitative evidence is currently unsupported by the manuscript as presented: the new dataset and the experimental protocol are not sufficiently described to allow verification of the headline result.

major comments (4)
  1. [Abstract / Dataset] The 'up to 12% (F1) improvement' hinges on a 'new diverse-geometry geospatial ER dataset', but the abstract provides no construction details: no labeling protocol, no inter-annotator agreement, no dataset statistics, and no leakage analysis (e.g., whether geometries and labels derive from the same source). Without this documentation, the benchmark's validity cannot be assessed, and the central claim is unverifiable. The point-only datasets cannot exercise the geometry encoder, so the new dataset is the sole basis for the claimed advantage. This is load-bearing and must be addressed.
  2. [Evaluation / Baselines] The abstract does not identify which 'existing methods' are compared against, how they are configured, or whether any geometry-aware baselines are included. If the baselines are point-collapsed and cannot represent polygon topology, the reported F1 gap may reflect baseline limitations rather than Omni's encoder strength. Additionally, the 'up to' phrasing suggests best-case selection without reporting variance, number of runs, or statistical significance. A rigorous experimental protocol with error bars and a full set of baselines is required.
  3. [Model / Ablation] No ablation separates the contribution of the omni-geometry encoder from the Attribute Affinity mechanism and the language model. Since point-only datasets cannot exercise the geometry encoder, it is unclear whether the 12% gain arises from geometry encoding or from the text-matching component. Without an ablation that isolates geometry (e.g., a point-collapsed version of Omni on the diverse-geometry dataset), the causal role of the proposed encoder is unestablished.
  4. [LLM Comparison] The LLM experiments are described only as 'prompting strategies and learning scenarios' with 'competitive results'. No model size, prompting template, in-context example count, or evaluation setup is given. This prevents the reader from judging whether the comparison is fair (e.g., whether LLMs have access to the same attributes and geometry representations) and whether the conclusion is robust. This is a major omission in the experimental section.
minor comments (3)
  1. [Abstract] Acronyms such as ER and POI are used without expansion; considering the abstract is read broadly, define them on first use.
  2. [Abstract] The phrase 'up to 12%' is ambiguous: specify the absolute F1 value and the exact comparison configuration, or replace with a range across datasets.
  3. [General] The abstract should briefly describe the dataset size and geometry types (points, lines, polygons counts) to give context for the claimed improvement.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the claims are empirical comparisons; the new dataset's validity is a benchmark concern, not a derivation-level circularity.

full rationale

The paper's central claims are empirical: Omni's geometry encoder and Attribute Affinity mechanism are evaluated against existing methods on point-only datasets and a newly introduced diverse-geometry dataset. There is no derivation chain in which an output is defined in terms of an input, no fitted parameter is renamed as a prediction, and no load-bearing self-citation is apparent from the provided text. The new diverse-geometry dataset could in principle favor Omni by construction, but that would be a benchmark-validity or data-leakage issue, not a circularity of the form where a result is equivalent to its assumptions by definition. The abstract makes no formal claims that reduce to earlier equations, and the reported 12% F1 improvement is an empirical outcome, not an algebraic consequence of the model's architecture. Consistent with the reviewing rules, concerns about dataset labeling, leakage, and representativeness belong to correctness risk, not to the circularity axis. Therefore the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No separate invented entities are listed. Omni's encoder and Attribute Affinity mechanism are the paper's proposed artefacts, not ad hoc postulates invoked to explain a result; they stand or fall with the experiments themselves. The free-parameters ledger records what is actually chosen by hand: the reported 'up to' configuration and the undisclosed training setup. The axioms list the load-bearing domain assumptions of the method and evaluation, all of which are abstract-level assertions.

free parameters (2)
  • Best-case 'up to 12%' configuration selection = unreported
    The abstract reports the maximum F1 gain without saying which configuration produced it or what the spread across configurations was; this is a hand-selected headline figure.
  • Model and training hyperparameters = unreported
    No architecture size, optimizer, learning regime, or training data details are given, so the gain cannot be attributed to the geometry encoder versus model capacity.
assumptions (4)
  • domain assumption Heterogeneous geometries can be embedded by one learned encoder without losing the spatial information needed for matching.
    Central premise of the Omni geometry encoder; the entire improvement claim depends on it (abstract).
  • domain assumption The new diverse-geometry dataset has accurate ground truth and representative difficulty.
    The 12% F1 result is measured on this dataset; label quality, task difficulty, and leakage are not verifiable from the abstract.
  • domain assumption Pre-trained transformer language models over individual attributes capture the textual similarity needed for place matching.
    Foundation of the Attribute Affinity mechanism (abstract); standard in the PLM-based ER literature but still an assumption about transfer.
  • domain assumption Baselines were tuned fairly and compared under equal conditions.
    The claimed improvement over 'existing methods' is only interpretable if baselines were not handicapped; no tuning protocol is visible in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Omni Geometry Representation Learning vs Large Language Models for Geospatial Entity Resolution." pith.science (2026). https://pith.science/paper/PBZ5RJJX

@misc{pith2026250806584,
  author       = {Pith},
  title        = {Pith review of: Omni Geometry Representation Learning vs Large Language Models for Geospatial Entity Resolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PBZ5RJJX}},
  note         = {Machine review of arXiv:2508.06584}
}
read the original abstract

The development, integration, and maintenance of geospatial databases rely heavily on efficient and accurate matching procedures of Geospatial Entity Resolution (ER). While resolution of points-of-interest (POIs) has been widely addressed, resolution of entities with diverse geometries has been largely overlooked. This is partly due to the lack of a uniform technique for embedding heterogeneous geometries seamlessly into a neural network framework. Existing neural approaches simplify complex geometries to a single point, resulting in significant loss of spatial information. To address this limitation, we propose Omni, a geospatial ER model featuring an omni-geometry encoder. This encoder is capable of embedding point, line, polyline, polygon, and multi-polygon geometries, enabling the model to capture the complex geospatial intricacies of the places being compared. Furthermore, Omni leverages transformer-based pre-trained language models over individual textual attributes of place records in an Attribute Affinity mechanism. The model is rigorously tested on existing point-only datasets and a new diverse-geometry geospatial ER dataset. Omni produces up to 12% (F1) improvement over existing methods. Furthermore, we test the potential of Large Language Models (LLMs) to conduct geospatial ER, experimenting with prompting strategies and learning scenarios, comparing the results of pre-trained language model-based methods with LLMs. Results indicate that LLMs show competitive results.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

72 extracted references · 58 canonical work pages

  1. [1]

    Gazetteer matching for natural features in switzerland

    Acheson, E., et al., 2017. Gazetteer matching for natural features in switzerland. In: Proceedings of the 11th Workshop on Geographic Information Retrieval. 1--2

  2. [2]

    Machine learning for cross-gazetteer matching of natural features

    Acheson, E., Volpi, M., and Purves, R.S., 2020. Machine learning for cross-gazetteer matching of natural features. International Journal of Geographical Information Science, 34 (4), 708--734

  3. [3]

    Gpt-4 technical report

    Achiam, J., et al., 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774

  4. [4]

    Assessment of the accuracy of geonames gazetteer data

    Ahlers, D., 2013. Assessment of the accuracy of geonames gazetteer data. In: Proceedings of the 7th workshop on geographic information retrieval. 74--81

  5. [5]

    Geospatial entity resolution

    Balsebre, P., et al., 2022. Geospatial entity resolution. In: Proceedings of the ACM Web Conference 2022. 3061--3070

  6. [6]

    Mining geospatial relationships from text

    Balsebre, P., et al., 2023. Mining geospatial relationships from text. Proceedings of the ACM on Management of Data, 1 (1), 1--26

  7. [7]

    Enriching word vectors with subword information

    Bojanowski, P., et al., 2017. Enriching word vectors with subword information. Transactions of the association for computational linguistics, 5, 135--146

  8. [8]

    Geometric deep learning: going beyond euclidean data

    Bronstein, M.M., et al., 2017. Geometric deep learning: going beyond euclidean data. IEEE Signal Processing Magazine, 34 (4), 18--42

Show all 72 references
  1. [9]

    and Stockinger, K., 2020

    Brunner, U. and Stockinger, K., 2020. Entity matching with transformer architectures-a step forward in data integration. In: 23rd International Conference on Extending Database Technology, Copenhagen, 30 March-2 April 2020. OpenProceedings, 463--473

  2. [10]

    Palm: Scaling language modeling with pathways

    Chowdhery, A., et al., 2023. Palm: Scaling language modeling with pathways. Journal of Machine Learning Research, 24 (240), 1--113

  3. [11]

    An overview of end-to-end entity resolution for big data

    Christophides, V., et al., 2020. An overview of end-to-end entity resolution for big data. ACM Computing Surveys (CSUR), 53 (6), 1--42

  4. [12]

    Geo-aware networks for fine-grained recognition

    Chu, G., et al., 2019. Geo-aware networks for fine-grained recognition. In: Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops. 247--254

  5. [13]

    and Barbosa, L., 2021

    Cousseau, V. and Barbosa, L., 2021. Linking place records using multi-view encoders. Neural Computing and Applications, 33 (18), 12103--12119

  6. [14]

    Qlora: Efficient finetuning of quantized llms

    Dettmers, T., et al., 2024. Qlora: Efficient finetuning of quantized llms. Advances in Neural Information Processing Systems, 36

  7. [15]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Devlin, J., et al., 2018. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805

  8. [16]

    A survey on in-context learning

    Dong, Q., et al., 2024. A survey on in-context learning. In: Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 1107--1128

  9. [17]

    and Peucker, T.K., 1973

    Douglas, D.H. and Peucker, T.K., 1973. Algorithms for the reduction of the number of points required to represent a digitized line or its caricature. Cartographica: the International Journal for Geographic Information and Geovisualization, 10 (2), 112--122

  10. [18]

    Cost-effective in-context learning for entity resolution: A design space exploration

    Fan, M., et al., 2024. Cost-effective in-context learning for entity resolution: A design space exploration. In: 2024 IEEE 40th International Conference on Data Engineering (ICDE). IEEE, 3696--3709

  11. [19]

    Constructing gazetteers from volunteered big geo-data based on hadoop

    Gao, S., et al., 2017. Constructing gazetteers from volunteered big geo-data based on hadoop. Computers, Environment and Urban Systems, 61, 172--186

  12. [20]

    Automated conflation of digital gazetteer data

    Hastings, J., 2008. Automated conflation of digital gazetteer data. International Journal of Geographical Information Science, 22 (10), 1109--1127

  13. [21]

    Deep residual learning for image recognition

    He, K., et al., 2016. Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 770--778

  14. [22]

    and Papadakis, G., 2024

    Kasinikos, I.I.A. and Papadakis, G., 2024. Entity resolution with small-scale llms: A study on prompting strategies and hardware limitations

  15. [23]

    Large language models are zero-shot reasoners

    Kojima, T., et al., 2022. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35, 22199--22213

  16. [24]

    Evaluation of entity resolution approaches on real-world match problems

    K\" o pcke, H., Thor, A., and Rahm, E., 2010. Evaluation of entity resolution approaches on real-world match problems. Proc. VLDB Endow., 3 (1–2), 484–493. ://dl.acm.org/doi/10.14778/1920841.1920904

  17. [25]

    Geographic ontologies, gazetteers and multilingualism

    Laurini, R., 2015. Geographic ontologies, gazetteers and multilingualism. Future Internet, 7 (1), 1--23

  18. [26]

    Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension

    Lewis, M., 2019. Bart: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. arXiv preprint arXiv:1910.13461

  19. [27]

    Booster: Leveraging large language models for enhancing entity resolution

    Li, H., et al., 2024. Booster: Leveraging large language models for enhancing entity resolution. In: Companion Proceedings of the ACM on Web Conference 2024. 1043--1046

  20. [28]

    Pointcnn: Convolution on x-transformed points

    Li, Y., et al., 2018. Pointcnn: Convolution on x-transformed points. Advances in Neural Information Processing Systems, 31

  21. [29]

    Deep entity matching with pre-trained language models

    Li, Y., et al., 2020. Deep entity matching with pre-trained language models. Proceedings of the VLDB Endowment, 14 (1), 50--60

  22. [30]

    Roberta: A robustly optimized bert pretraining approach

    Liu, Y., et al., 2019. Roberta: A robustly optimized bert pretraining approach. ArXiv, abs/1907.11692. ://api.semanticscholar.org/CorpusID:198953378

  23. [31]

    Presence-only geographical priors for fine-grained image classification

    Mac Aodha, O., Cole, E., and Perona, P., 2019. Presence-only geographical priors for fine-grained image classification. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. 9596--9606

  24. [32]

    On the opportunities and challenges of foundation models for geospatial artificial intelligence

    Mai, G., et al., 2023 a . On the opportunities and challenges of foundation models for geospatial artificial intelligence. arXiv preprint arXiv:2304.06798

  25. [33]

    A review of location encoding for geoai: methods and applications

    Mai, G., et al., 2022. A review of location encoding for geoai: methods and applications. International Journal of Geographical Information Science, 36 (4), 639--673

  26. [34]

    Multi-scale representation learning for spatial feature distributions using grid cells

    Mai, G., et al., 2020. Multi-scale representation learning for spatial feature distributions using grid cells. arXiv preprint arXiv:2003.00824

  27. [35]

    Towards general-purpose representation learning of polygonal geometries

    Mai, G., et al., 2023 b . Towards general-purpose representation learning of polygonal geometries. GeoInformatica, 27 (2), 289--340

  28. [36]

    A supervised machine learning approach for duplicate detection over gazetteer records

    Martins, b., 2011. A supervised machine learning approach for duplicate detection over gazetteer records. In: International Conference on GeoSpatial Sematics. Springer, 34--51

  29. [37]

    Weighted multi-attribute matching of user-generated points of interest

    McKenzie, G., Janowicz, K., and Adams, B., 2013. Weighted multi-attribute matching of user-generated points of interest. In: Proceedings of the 21st ACM SIGSPATIAL international conference on advances in geographic information systems. 440--443

  30. [38]

    Distributed representations of words and phrases and their compositionality

    Mikolov, T., et al., 2013. Distributed representations of words and phrases and their compositionality. Advances in Neural Information Processing Systems, 26

  31. [39]

    Can foundation models wrangle your data? Proceedings of the VLDB Endowment, 16 (4), 738--746

    Narayan, A., et al., 2022. Can foundation models wrangle your data? Proceedings of the VLDB Endowment, 16 (4), 738--746

  32. [40]

    A multi-facet analysis of bert-based entity matching models

    Paganelli, M., Tiano, D., and Guerra, F., 2023. A multi-facet analysis of bert-based entity matching models. The VLDB Journal, 1--26

  33. [41]

    and Bizer, C., 2021

    Peeters, R. and Bizer, C., 2021. Dual-objective fine-tuning of bert for entity matching. Proceedings of the VLDB Endowment, 14, 1913--1921

  34. [42]

    Entity matching using large language models

    Peeters, R., Steiner, A., and Bizer, C., 2023. Entity matching using large language models. arXiv preprint arXiv:2310.11244

  35. [43]

    Pointnet: Deep learning on point sets for 3d classification and segmentation

    Qi, C.R., et al., 2017. Pointnet: Deep learning on point sets for 3d classification and segmentation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 652--660

  36. [44]

    A critical evaluation of location based services and their potential

    Raper, J., et al., 2007. A critical evaluation of location based services and their potential. Journal of Location Based Services, 1 (1), 5--45

  37. [45]

    Sentence-bert: Sentence embeddings using siamese bert-networks

    Reimers, N., 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084

  38. [46]

    Toponym matching through deep neural networks

    Santos, R., et al., 2018. Toponym matching through deep neural networks. International Journal of Geographical Information Science, 32 (2), 324--348

  39. [47]

    Entity resolution in geospatial data integration

    Sehgal, V., Getoor, L., and Viechnicki, P.D., 2006. Entity resolution in geospatial data integration. In: Proceedings of the 14th Annual ACM International Symposium on Advances in Geographic Information Systems. 83--90

  40. [48]

    Multi-source toponym data integration and mediation for a meta-gazetteer service

    Smart, P.D., Jones, C.B., and Twaroch, F.A., 2010. Multi-source toponym data integration and mediation for a meta-gazetteer service. In: Proceedings of the International Conference on Geographic Information Science. Springer, 234--248

  41. [49]

    Conflating point of interest (poi) data: A systematic review of matching methods

    Sun, K., et al., 2023. Conflating point of interest (poi) data: A systematic review of matching methods. Computers, Environment and Urban Systems, 103, 101977

  42. [50]

    Improving image classification with location context

    Tang, K., et al., 2015. Improving image classification with location context. In: Proceedings of the IEEE International Conference on Computer Vision. 1008--1016

  43. [51]

    Llama 2: Open foundation and fine-tuned chat models

    Touvron, H., et al., 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288

  44. [52]

    Learning localized generative models for 3d point clouds via graph convolution

    Valsesia, D., Fracastoro, G., and Magli, E., 2018. Learning localized generative models for 3d point clouds via graph convolution. In: International Conference on Learning Representations

  45. [53]

    Attention is all you need

    Vaswani, A., et al., 2017. Attention is all you need. Advances in Neural Information Processing Systems, 30

  46. [54]

    Deep learning for classification tasks on geospatial vector polygons

    Veer, R.v., Bloem, P., and Folmer, E., 2018. Deep learning for classification tasks on geospatial vector polygons. arXiv preprint arXiv:1806.03857

  47. [55]

    Crowder: crowdsourcing entity resolution

    Wang, J., et al., 2012. Crowder: crowdsourcing entity resolution. Proc. VLDB Endow., 5 (11), 1483–1494. ://dl.acm.org/doi/10.14778/2350229.2350263

  48. [56]

    Improving text embeddings with large language models

    Wang, L., et al., 2024 a . Improving text embeddings with large language models. In: L.W. Ku, A. Martins and V. Srikumar, eds. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), August, Bangkok, Thailand. Associatio...

  49. [57]

    Gpt-ner: Named entity recognition via large language models

    Wang, S., et al., 2023. Gpt-ner: Named entity recognition via large language models. arXiv preprint arXiv:2304.10428

  50. [58]

    Match, compare, or select? an investigation of large language models for entity matching

    Wang, T., et al., 2024 b . Match, compare, or select? an investigation of large language models for entity matching. arXiv preprint arXiv:2405.16884

  51. [59]

    Encoding crowd interaction with deep neural network for pedestrian trajectory prediction

    Xu, Y., Piao, Z., and Gao, S., 2018. Encoding crowd interaction with deep neural network for pedestrian trajectory prediction. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 5275--5284

  52. [60]

    Graph convolutional autoencoder model for the shape coding and cognition of buildings in maps

    Yan, X., et al., 2021. Graph convolutional autoencoder model for the shape coding and cognition of buildings in maps. International Journal of Geographical Information Science, 35 (3), 490--512

  53. [61]

    Place deduplication with embeddings

    Yang, C., et al., 2019. Place deduplication with embeddings. In: The World Wide Web Conference. 3420--3426

  54. [62]

    Gps2vec: Towards generating worldwide gps embeddings

    Yin, Y., et al., 2019. Gps2vec: Towards generating worldwide gps embeddings. In: Proceedings of the 27th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems. 416--419

  55. [63]

    Pre-trained embeddings for entity resolution: an experimental analysis

    Zeakis, A., et al., 2023. Pre-trained embeddings for entity resolution: an experimental analysis. Proceedings of the VLDB Endowment, 16 (9), 2225--2238

  56. [64]

    Sr-lstm: State refinement for lstm towards pedestrian trajectory prediction

    Zhang, P., et al., 2019. Sr-lstm: State refinement for lstm towards pedestrian trajectory prediction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 12085--12094

  57. [65]

    Detecting nearly duplicated records in location datasets

    Zheng, Y., et al., 2010. Detecting nearly duplicated records in location datasets. In: Proceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic Information Systems. 137--143

  58. [66]

    A points of interest matching method using a multivariate weighting function with gradient descent optimization

    Zhou, Y., et al., 2021. A points of interest matching method using a multivariate weighting function with gradient descent optimization. Transactions in GIS, 25 (1), 359--381

  59. [67]

    Autotqa: Towards autonomous tabular question answering through multi-agent large language models

    Zhu, J.P., et al., 2024. Autotqa: Towards autonomous tabular question answering through multi-agent large language models. Proceedings of the VLDB Endowment, 17 (12), 3920--3933

  60. [68]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should ...

  61. [69]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@firs...

  62. [70]

    Ty @ @ _@f @ @ @ @

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibset...

  63. [71]

    , " * write output.state after.block = add.period

    ENTRY address archive author booktitle chapter collaboration edition editor eid howpublished institution journal key lastchecked month note number numpages organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS ou...

  64. [72]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.