arxiv: 2604.07956 · v2 · submitted 2026-04-09 · 💻 cs.AI

Recognition: 2 theorem links

· Lean Theorem

MONETA: Multimodal Industry Classification through Geographic Information with Multi Agent Systems

Arda Y\"uksel , Gabriel Thiem , Susanne Walter , Patrick Felka , Gabriela Alves Werb , Ivan Habernal

Authors on Pith no claims yet

Pith reviewed 2026-05-10 18:25 UTC · model grok-4.3

classification 💻 cs.AI

keywords industry classificationmultimodal learninglarge language modelsgeospatial dataNACE codesbenchmark datasetsatellite imagerymulti-turn prompting

0 comments

The pith

Multimodal models assign NACE industry labels to companies using websites, maps, and satellite images without training data.

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

The paper creates MONETA, a benchmark of 1,000 European companies labeled with 20 NACE economic activity codes, to test whether multimodal large language models can replicate expert classification. It combines textual inputs from websites, Wikipedia, and Wikidata with geospatial inputs from OpenStreetMap and satellite imagery. The work matters because manual labeling of company registers is costly and retraining models for each scheme update is impractical. Training-free baselines reach 62.10 percent with open models and 74.10 percent with closed models; multi-turn prompting, added context, and required explanations raise results by as much as 22.80 percent.

Core claim

MONETA is the first multimodal benchmark for industry classification that pairs text sources (company websites, Wikipedia, Wikidata) with geospatial sources (OpenStreetMap, satellite imagery) across 1,000 European businesses and 20 NACE labels. A training-free baseline using multimodal large language models achieves 62.10 percent accuracy with open-source models and 74.10 percent with closed-source models. Accuracy rises by up to 22.80 percent when multi-turn design, context enrichment, and classification explanations are combined.

What carries the argument

The MONETA benchmark dataset together with multi-turn prompting of multimodal large language models that jointly process text and image inputs to predict NACE codes.

If this is right

Existing public text and geospatial data can substitute for manual expert verification on industry classification tasks.
Performance gains come from interaction design rather than model retraining or new labels.
The released dataset and guidelines support testing of future multimodal models on the same task.
The approach scales to updates in classification schemes without repeating large data collection efforts.
Geospatial signals complement text when company descriptions are sparse or ambiguous.

Where Pith is reading between the lines

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

If the same signals work outside Europe, public registers could adopt automated labeling at lower cost.
The same multimodal setup could extend to product or risk classification where location data is available.
Satellite imagery may add value mainly for companies whose websites or Wikipedia entries are minimal.
Errors that persist after multi-turn prompting would point to gaps in current open geospatial data coverage.

Load-bearing premise

Readily available multimodal resources already contain enough accurate signals to match expert NACE labels without new labeled data collection.

What would settle it

A manual re-audit of the 1,000 companies in which expert NACE labels diverge from the best multi-turn model outputs on more than 30 percent of cases.

Figures

Figures reproduced from arXiv: 2604.07956 by Arda Y\"uksel, Gabriela Alves Werb, Gabriel Thiem, Ivan Habernal, Patrick Felka, Susanne Walter.

**Figure 1.** Figure 1: Zero-Shot vs Multi-Turn comparison for MONETA. (1) Zero-shot pipeline (left): Available resources are forwarded into the industry classifier MLLM together. Explanations and classifications are obtained. (2) Multi-Turn pipeline (right): Each resource is processed by separate specialized agents. Intermediate clues from these agents are processed by the decisionmaking agent, returning explanation and class… view at source ↗

**Figure 2.** Figure 2: Overview of the dataset preparation process. (1) NACE section XMLs are initially converted to the OSM [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Figure 3: Example with ground truth NACE sections K and prediction G. Clues and predictions are obtained via [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Baseline unknown ratio of InternVL 3 for [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: InternVL 3 (8B and 38B) correctness and ef [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

**Figure 6.** Figure 6: Average number of associated resources (OSM images, satellite images, website text, Wikidata, and [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗

**Figure 7.** Figure 7: Spatial distribution of MONETA entries across Europe, showing the geographic coverage of OSM-derived entities used for NACE classification [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗

**Figure 8.** Figure 8: Distribution of OpenStreetMap (OSM) tags across NACE economic activity sections. The horizontal bar [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗

**Figure 9.** Figure 9: RDF/XML NACE official guideline content for Section K (Financial and Insurance Activities). [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗

**Figure 11.** Figure 11: NACE section distribution for MONETA-10K. Sections less than 2% are grouped into other for demonstration. et al., 2022). C.2 Infrastructure In our experiments, we used NVIDIA A100 40GB GPUs and increased number of GPUs depending on the model size. C.3 Hyperparameters During the ablation studies for the text only benchmarks, we fine tuned models with LoRA (Hu et al., [PITH_FULL_IMAGE:figures/full_fig_p0… view at source ↗

**Figure 12.** Figure 12: NACE Section-wise analysis for InternVL 2.5 (4B and 38B), InternVL 3 (8B and 38B), and QwenVL 2.5 [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗

**Figure 13.** Figure 13: Experiment configuration results given in Accuracy, Precision, and Recall for InternVL 2.5 (4B and [PITH_FULL_IMAGE:figures/full_fig_p020_13.png] view at source ↗

**Figure 14.** Figure 14: Clue Keywords confusion matrices obtained for InternVL 3 (8B and 38B), and QwenVL 2.5 (7B and [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗

read the original abstract

Industry classification schemes are integral parts of public and corporate databases as they classify businesses based on economic activity. Due to the size of the company registers, manual annotation is costly, and fine-tuning models with every update in industry classification schemes requires significant data collection. We replicate the manual expert verification by using existing or easily retrievable multimodal resources for industry classification. We present MONETA, the first multimodal industry classification benchmark with text (Website, Wikipedia, Wikidata) and geospatial sources (OpenStreetMap and satellite imagery). Our dataset enlists 1,000 businesses in Europe with 20 economic activity labels according to EU guidelines (NACE). Our training-free baseline reaches 62.10% and 74.10% with open and closed-source Multimodal Large Language Models (MLLM). We observe an increase of up to 22.80% with the combination of multi-turn design, context enrichment, and classification explanations. We will release our dataset and the enhanced guidelines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

MONETA builds a new 1000-example benchmark for NACE classification from websites, wikis, OSM, and satellite images, then shows that multi-turn MLLM agents lift accuracy by up to 22.8% over single-pass use.

read the letter

MONETA puts together a dataset of 1000 European companies labeled with 20 NACE codes and pairs each with readily available text and geospatial sources. The main finding is that off-the-shelf MLLMs reach 62% and 74% accuracy in a training-free setup, with a multi-turn agent workflow plus context enrichment and required explanations adding up to a 22.8% relative gain. They plan to release the data and guidelines, which is the most useful part for anyone maintaining company registers. The practical framing is honest: manual labeling is expensive and schemes change, so leveraging public multimodal signals makes sense for this narrow task. The agent comparison is concrete and worth noting for people experimenting with structured prompting. The soft spots are the missing pieces around ground truth and controls. There is no report on how the 1000 labels were produced or any inter-annotator agreement, so the accuracy numbers sit on unverified expert annotations. No per-modality ablations appear, which leaves open whether the geospatial sources actually move the needle or if text mentions of industry do most of the work. The gains also lack any statistical test, so it is unclear how much is real improvement versus prompt sensitivity. The scope stays limited to Europe and 20 classes, which is fine for a benchmark but restricts broader claims. This is for applied researchers working on business databases or multimodal agents on real classification problems. It deserves peer review because the dataset is new and the empirical comparison is straightforward, even if the authors need to add the missing checks on label quality and modality value.

Referee Report

3 major / 3 minor

Summary. The manuscript introduces MONETA, the first multimodal benchmark for NACE industry classification comprising 1,000 European businesses annotated with 20 economic activity classes. It uses text sources (company websites, Wikipedia, Wikidata) and geospatial data (OpenStreetMap, satellite imagery) in a training-free setting with open- and closed-source MLLMs, reporting baseline accuracies of 62.10% and 74.10% respectively, and relative gains of up to 22.80% from multi-turn design, context enrichment, and explanatory prompts. The central claim is that these readily available multimodal resources suffice to replicate manual expert verification without new labeled data collection.

Significance. If validated, MONETA would supply a reusable benchmark and scalable, annotation-light approach to industry classification, directly addressing the cost of maintaining large company registers under evolving NACE schemes. The planned release of the dataset and enhanced guidelines strengthens its utility for the community. The work's empirical focus on real-world multimodal signals is timely for applied AI in economics, though its impact hinges on demonstrating that performance derives from genuine multimodal integration rather than text alone.

major comments (3)

[Dataset Construction] Dataset section: The 1,000 NACE labels are presented as expert annotations replicating manual verification, yet no details are given on label provenance, annotation protocol, or inter-annotator agreement. This information is required to substantiate the claim that the multimodal pipeline matches expert performance.
[Results] Results and Experiments sections: The 22.80% lift from multi-turn, context enrichment, and explanations is reported without modality ablations or per-source contribution analysis. It remains unclear whether geospatial inputs (OSM and satellite imagery) add discriminative power beyond textual mentions of industry activity, which is load-bearing for the multimodal premise.
[Experiments] Evaluation Methodology: No data splits, statistical significance tests for accuracy differences, or controls for prompt sensitivity are described. These omissions prevent assessment of whether the reported baselines and gains are robust or sensitive to implementation choices.

minor comments (3)

[Introduction] The abstract and introduction should explicitly cite prior work on automated NACE or SIC classification to better position the novelty of the multimodal benchmark.
[Method] Figure captions and the multi-agent system diagram would benefit from clearer indication of which agents handle which modalities and how outputs are aggregated.
Minor inconsistencies in terminology (e.g., 'multi-turn' vs. 'multi-turn design') and occasional missing units in accuracy tables should be standardized.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We address each major point below and commit to revisions that will strengthen the presentation of MONETA while preserving the core contributions of the work.

read point-by-point responses

Referee: [Dataset Construction] Dataset section: The 1,000 NACE labels are presented as expert annotations replicating manual verification, yet no details are given on label provenance, annotation protocol, or inter-annotator agreement. This information is required to substantiate the claim that the multimodal pipeline matches expert performance.

Authors: We agree that explicit details on label provenance and protocol are needed to support the claim. The NACE labels were obtained directly from official European business registries (cross-referenced with public company databases) to replicate standard expert verification without new annotation. We will add a subsection in the Dataset section describing the exact sources, the alignment protocol with NACE Rev. 2 guidelines, and clarification that traditional inter-annotator agreement does not apply because labels derive from authoritative registers rather than multiple independent annotators. These changes will appear in the revised manuscript. revision: yes
Referee: [Results] Results and Experiments sections: The 22.80% lift from multi-turn, context enrichment, and explanations is reported without modality ablations or per-source contribution analysis. It remains unclear whether geospatial inputs (OSM and satellite imagery) add discriminative power beyond textual mentions of industry activity, which is load-bearing for the multimodal premise.

Authors: We acknowledge that modality-specific ablations are required to isolate the contribution of geospatial data. Although the reported gains reflect the full multimodal pipeline, we will add ablation experiments that systematically remove OpenStreetMap and satellite imagery while retaining text sources, together with a per-source contribution breakdown. These results will be inserted into the Results and Experiments sections to demonstrate that geospatial inputs provide measurable discriminative value beyond text alone. revision: yes
Referee: [Experiments] Evaluation Methodology: No data splits, statistical significance tests for accuracy differences, or controls for prompt sensitivity are described. These omissions prevent assessment of whether the reported baselines and gains are robust or sensitive to implementation choices.

Authors: Because the approach is training-free and evaluates the full set of 1,000 samples in a zero-shot regime, conventional data splits are not applicable. To improve methodological rigor, we will incorporate statistical significance tests (McNemar’s test) for all reported accuracy differences and add a prompt-sensitivity analysis that varies phrasing of the classification and explanation prompts while reporting result variance. These elements will be added to the Evaluation Methodology subsection of the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical benchmark with independent evaluation results

full rationale

The paper introduces a new multimodal benchmark dataset (MONETA) of 1,000 European businesses labeled with NACE codes and reports direct accuracy numbers from training-free MLLM evaluations on text and geospatial inputs. No equations, derivations, fitted parameters, or self-citation chains are present that would reduce the reported accuracies (62.10%, 74.10%, or the 22.80% gains) to quantities defined inside the study by construction. The central claims rest on the assembled dataset and the observed model outputs rather than any self-referential step; this is a standard empirical benchmark paper whose results are falsifiable against the released data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No mathematical model or derivation is present; the study relies on off-the-shelf MLLMs and publicly available data sources.

pith-pipeline@v0.9.0 · 5481 in / 1164 out tokens · 59653 ms · 2026-05-10T18:25:07.400017+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We propose, MONETA, a multimodal industry classification benchmark... Zero-Shot and Multi-Turn pipelines... frequency vectors... Correctness and Effectiveness metrics.
IndisputableMonolith/Foundation/AlexanderDuality.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

NACE to OSM mapping... satellite imagery via ESRI REST API... 20 NACE sections.

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

50 extracted references · 34 canonical work pages · 5 internal anchors

[1]

Carole A Ambler and James E Kristoff. 1998. https://doi.org/10.1016/S0740-624X(98)90003-X Introducing the North American industry classification system . Government Information Quarterly, 15(3):263--273

work page doi:10.1016/s0740-624x(98)90003-x 1998
[2]

Matteo Ambrois, Vincenzo Butticè, Federico Caviggioli, Giovanni Cerulli, Annalisa Croce, Antonio De Marco, Andrea Giordano, Giuliano Resce, Laura Toschi, Elisa Ughetto, and Antonio Zinilli. 2023. https://EconPapers.repec.org/RePEc:zbw:eifwps:202391 Using machine learning to map the european cleantech sector . EIF Working Paper Series 2023/91, European Inv...

2023
[3]

Freeman, J

Madeline Loui Anderson, Miriam Cha, William T. Freeman, J. Taylor Perron, Nathaniel Maidel, and Kerri Cahoy. 2025. https://openreview.net/forum?id=fEMTiazivQ Measuring and mitigating hallucinations in vision-language dataset generation for remote sensing . In Workshop on Preparing Good Data for Generative AI: Challenges and Approaches

2025
[4]

Jinze Bai, Shuai Bai, Shusheng Yang, Shijie Wang, Sinan Tan, Peng Wang, Junyang Lin, Chang Zhou, and Jingren Zhou. 2023. Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond. arXiv preprint arXiv:2308.12966

work page internal anchor Pith review Pith/arXiv arXiv 2023
[5]

Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, and 8 others. 2025. Qwen2.5-vl technical report. arXiv preprint arXiv:2502.13923

work page internal anchor Pith review Pith/arXiv arXiv 2025
[6]

Hannah Béchara, Ran Zhang, Shuzhou Yuan, and Slava Jankin. 2022. https://doi.org/10.1109/BigData55660.2022.10020787 Applying NLP Techniques to Classify Businesses by their International Standard Industrial Classification ( ISIC ) Code . In 2022 IEEE International Conference on Big Data ( Big Data ) , pages 3472--3477

work page doi:10.1109/bigdata55660.2022.10020787 2022
[7]

Wei Chen, Xixuan Hao, Yuankai Wu, and Yuxuan Liang. 2024 a . https://doi.org/10.52202/079017-2121 Terra: A multimodal spatio-temporal dataset spanning the earth . In Advances in Neural Information Processing Systems, volume 37, pages 66329--66356. Curran Associates, Inc

work page doi:10.52202/079017-2121 2024
[8]

Yuzhou Chen, Jiue-An Yang, Hugo Kyo Lee, Calvin Tribby, Tarik Benmarhnia, Marta Jankowska, and Yulia R. Gel. 2025. https://doi.org/10.1109/ICASSP49660.2025.10889624 Fusing Multimodality of Large Language Models and Satellite Imagery via Simplicial Contrastive Learning for Latent Urban Feature Identification and Environmental Application . In ICASSP 2025 -...

work page doi:10.1109/icassp49660.2025.10889624 2025
[9]

Zhe Chen, Weiyun Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Erfei Cui, Jinguo Zhu, Shenglong Ye, Hao Tian, Zhaoyang Liu, and 1 others. 2024 b . Expanding performance boundaries of open-source multimodal models with model, data, and test-time scaling. arXiv preprint arXiv:2412.05271

work page internal anchor Pith review Pith/arXiv arXiv 2024
[10]

Annalisa Croce, Laura Toschi, Elisa Ughetto, and Sara Zanni. 2024. https://doi.org/10.1016/j.enpol.2024.114006 Cleantech and policy framework in Europe : A machine learning approach . Energy Policy, 186:114006

work page doi:10.1016/j.enpol.2024.114006 2024
[11]

Michael Han Daniel Han and Unsloth team. 2023. http://github.com/unslothai/unsloth Unsloth

2023
[12]

Guy Stephane Waffo Dzuyo, Gaël Guibon, Christophe Cerisara, and Luis Belmar-Letelier. 2025. https://doi.org/10.1609/aaai.v39i16.33806 Linking Industry Sectors and Financial Statements : A Hybrid Approach for Company Classification . Proceedings of the AAAI Conference on Artificial Intelligence, 39(16):16444--16452. Number: 16

work page doi:10.1609/aaai.v39i16.33806 2025
[13]

European Commission , editor. 2008. NACE Rev . 2: statistical classification of economic activities in the European Community . Publications Office, Luxembourg

2008
[14]

Thomas Faria and Tom Seimandi. 2023. Classifying companies in france using machine learning

2023
[15]

Gemini 2.5 Team . 2025. https://arxiv.org/abs/2507.06261 Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities . Preprint, arXiv:2507.06261

work page internal anchor Pith review Pith/arXiv arXiv 2025
[16]

Sucharita Gopal and Josh Pitts. 2024. https://doi.org/10.1007/978-3-031-74418-1_6 Geospatial Finance: Foundations and Applications , pages 225--273. Springer Nature Switzerland, Cham

work page doi:10.1007/978-3-031-74418-1_6 2024
[17]

Yanming Guo, Xiao Qian, Kevin Credit, and Jin Ma. 2025. https://doi.org/10.48550/arXiv.2502.06874 Group Reasoning Emission Estimation Networks . arXiv preprint. ArXiv:2502.06874 [cs]

work page doi:10.48550/arxiv.2502.06874 2025
[18]

Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. https://openreview.net/forum?id=nZeVKeeFYf9 Lo RA : Low-rank adaptation of large language models . In International Conference on Learning Representations

2022
[19]

Timotej Jagrič and Aljaž Herman. 2024. https://doi.org/10.3390/info15020089 AI Model for Industry Classification Based on Website Data . Information, 15(2):89. Number: 2 Publisher: Multidisciplinary Digital Publishing Institute

work page doi:10.3390/info15020089 2024
[20]

Pu Jin, Gui-Song Xia, Fan Hu, Qikai Lu, and Liangpei Zhang. 2018. https://doi.org/10.1109/IGARSS.2018.8518882 AID ++: An Updated Version of AID on Scene Classification . In IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium , pages 4721--4724. ISSN: 2153-7003

work page doi:10.1109/igarss.2018.8518882 2018
[21]

Sohail Ahmed Khan, Laurence Dierickx, Jan-Gunnar Furuly, Henrik Brattli Vold, Rano Tahseen, Carl-Gustav Linden, and Duc-Tien Dang-Nguyen. 2025. https://doi.org/10.1002/asi.24970 Debunking war information disorder: A case study in assessing the use of multimedia verification tools . Journal of the Association for Information Science and Technology, 76(5):752--769

work page doi:10.1002/asi.24970 2025
[22]

Heidi Kühnemann, Arnout van Delden, and Dick Windmeijer. 2020. https://doi.org/10.3233/SJI-200675 Exploring a knowledge-based approach to predicting nace codes of enterprises based on web page texts . Statistical Journal of the IAOS, 36(3):807--821

work page doi:10.3233/sji-200675 2020
[23]

Haifeng Li, Hao Jiang, Xin Gu, Jian Peng, Wenbo Li, Liang Hong, and Chao Tao. 2020. https://doi.org/10.3390/s20041226 CLRS : Continual Learning Benchmark for Remote Sensing Image Scene Classification . Sensors, 20(4):1226. Number: 4 Publisher: Multidisciplinary Digital Publishing Institute

work page doi:10.3390/s20041226 2020
[24]

Chu, Lujun Li, and Porawit Kamnoedboon

Qiang Li, Mingkun Tan, Xun Zhao, Dan Zhang, Daoan Zhang, Shengzhao Lei, Anderson S. Chu, Lujun Li, and Porawit Kamnoedboon. 2025. https://doi.org/10.18653/v1/2025.naacl-industry.4 How LLM s react to industrial spatio-temporal data? assessing hallucination with a novel traffic incident benchmark dataset . In Proceedings of the 2025 Conference of the Nation...

work page doi:10.18653/v1/2025.naacl-industry.4 2025
[25]

Haotian Liu, Chunyuan Li, Yuheng Li, and Yong Jae Lee. 2023 a . Improved baselines with visual instruction tuning

2023
[26]

Haotian Liu, Chunyuan Li, Yuheng Li, Bo Li, Yuanhan Zhang, Sheng Shen, and Yong Jae Lee. 2024. https://llava-vl.github.io/blog/2024-01-30-llava-next/ Llava-next: Improved reasoning, ocr, and world knowledge

2024
[27]

Haotian Liu, Chunyuan Li, Qingyang Wu, and Yong Jae Lee. 2023 b . Visual instruction tuning

2023
[28]

Ivan Malashin, Igor Masich, Vadim Tynchenko, Vladimir Nelyub, Aleksei Borodulin, and Andrei Gantimurov. 2024. https://doi.org/10.3390/bdcc8060068 Application of Natural Language Processing and Genetic Algorithm to Fine - Tune Hyperparameters of Classifiers for Economic Activities Analysis . Big Data and Cognitive Computing, 8(6):68. Number: 6 Publisher: M...

work page doi:10.3390/bdcc8060068 2024
[29]

Ali Mansourian and Rachid Oucheikh. 2024. https://doi.org/10.3390/ijgi13100348 ChatGeoAI : Enabling Geospatial Analysis for Public through Natural Language , with Large Language Models . ISPRS International Journal of Geo-Information, 13(10):348

work page doi:10.3390/ijgi13100348 2024
[30]

Ethan Mendes, Yang Chen, James Hays, Sauvik Das, Wei Xu, and Alan Ritter. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.957 Granular Privacy Control for Geolocation with Vision Language Models . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages 17240--17292, Miami, Florida, USA. Association for Computati...

work page doi:10.18653/v1/2024.emnlp-main.957 2024
[31]

Hibiki Nakatani, Hiroki Teranishi, Shohei Higashiyama, Yuya Sawada, Hiroki Ouchi, and Taro Watanabe. 2025. https://aclanthology.org/2025.coling-main.486/ A Text Embedding Model with Contrastive Example Mining for Point -of- Interest Geocoding . In Proceedings of the 31st International Conference on Computational Linguistics , pages 7279--7291, Abu Dhabi, ...

2025
[32]

Sanaz Saki Norouzi and Pascal Hitzler. 2025. https://doi.org/10.1609/aaaiss.v5i1.35609 Knowledge- Enhanced Geospatial QA : Integrating Wikidata Fact Verification with LLMs . Proceedings of the AAAI Symposium Series, 5(1):334--339

work page doi:10.1609/aaaiss.v5i1.35609 2025
[33]

OpenAI . 2025. Gpt-5. https://openai.com. Large language model

2025
[34]

Maryan Rizinski, Andrej Jankov, Vignesh Sankaradas, Eugene Pinsky, Igor Miskovski, and Dimitar Trajanov. 2023. https://doi.org/10.48550/arXiv.2305.01028 Company classification using zero-shot learning . arXiv preprint. ArXiv:2305.01028 [cs] version: 2

work page doi:10.48550/arxiv.2305.01028 2023
[35]

Mohamad Hakam Shams Eddin and J\" u rgen Gall. 2024. https://proceedings.neurips.cc/paper_files/paper/2024/file/aa7259c82d642e47d5661f3218cdcad2-Paper-Conference.pdf Identifying spatio-temporal drivers of extreme events . In Advances in Neural Information Processing Systems, volume 37, pages 93714--93766. Curran Associates, Inc

2024
[36]

Zirui Song, Jingpu Yang, Yuan Huang, Jonathan Tonglet, Zeyu Zhang, Tao Cheng, Meng Fang, Iryna Gurevych, and Xiuying Chen. 2025. https://doi.org/10.48550/arXiv.2502.13759 Geolocation with Real Human Gameplay Data : A Large - Scale Dataset and Human - Like Reasoning Framework . arXiv preprint. ArXiv:2502.13759 [cs]

work page doi:10.48550/arxiv.2502.13759 2025
[37]

Yuan Tao, Wanzeng Liu, Jun Chen, Jingxiang Gao, Ran Li, Xinpeng Wang, Ye Zhang, Jiaxin Ren, Shunxi Yin, Xiuli Zhu, Tingting Zhao, Xi Zhai, and Yunlu Peng. 2025. https://doi.org/10.1016/j.jag.2024.104353 A graph-based multimodal data fusion framework for identifying urban functional zone . International Journal of Applied Earth Observation and Geoinformati...

work page doi:10.1016/j.jag.2024.104353 2025
[38]

UN . 2008. https://doi.org/10.18356/8722852c-en International Standard Industrial Classification of All Economic Activities ( ISIC ), Rev .4 . Statistical Papers ( Ser . M ). United Nations, s.l

work page doi:10.18356/8722852c-en 2008
[39]

Dimitrios Vamvourellis, Máté Tóth, Snigdha Bhagat, Dhruv Desai, Dhagash Mehta, and Stefano Pasquali. 2024. https://doi.org/10.1109/CIFEr62890.2024.10772990 Company similarity using large language models . In 2024 IEEE Symposium on Computational Intelligence for Financial Engineering and Economics (CIFEr), pages 1--9

work page doi:10.1109/cifer62890.2024.10772990 2024
[40]

Peng Wang, Shuai Bai, Sinan Tan, Shijie Wang, Zhihao Fan, Jinze Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Yang Fan, Kai Dang, Mengfei Du, Xuancheng Ren, Rui Men, Dayiheng Liu, Chang Zhou, Jingren Zhou, and Junyang Lin. 2024 a . Qwen2-vl: Enhancing vision-language model's perception of the world at any resolution. arXiv preprint arXiv:2409.12191

work page internal anchor Pith review Pith/arXiv arXiv 2024
[41]

Weiyun Wang, Zhe Chen, Wenhai Wang, Yue Cao, Yangzhou Liu, Zhangwei Gao, Jinguo Zhu, Xizhou Zhu, Lewei Lu, Yu Qiao, and Jifeng Dai. 2024 b . Enhancing the reasoning ability of multimodal large language models via mixed preference optimization. arXiv preprint arXiv:2411.10442

work page arXiv 2024
[42]

Gabriela Alves Werb, Patrick Felka, Lisa Reichenbach, Susanne Walter, and Ece Yalcin-Roder. 2024. https://doi.org/10.1109/BigData62323.2024.10825029 Geospatial Data and Multimodal Fact - Checking for Validating Company Data . In 2024 IEEE International Conference on Big Data ( BigData ) , pages 3329--3332. ISSN: 2573-2978

work page doi:10.1109/bigdata62323.2024.10825029 2024
[43]

Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, and 3 others. 2020. https://www.aclweb.org/anthology/2020.emnlp-demos.6 Transformers...

2020
[44]

Gui-Song Xia, Jingwen Hu, Fan Hu, Baoguang Shi, Xiang Bai, Yanfei Zhong, Liangpei Zhang, and Xiaoqiang Lu. 2017. https://doi.org/10.1109/TGRS.2017.2685945 AID : A Benchmark Data Set for Performance Evaluation of Aerial Scene Classification . IEEE Transactions on Geoscience and Remote Sensing, 55(7):3965--3981

work page doi:10.1109/tgrs.2017.2685945 2017
[45]

Wenjia Xu, Zijian Yu, Yixu Wang, Jiuniu Wang, and Mugen Peng. 2024. https://doi.org/10.48550/arXiv.2406.07089 RS - Agent : Automating Remote Sensing Tasks through Intelligent Agents . arXiv preprint. ArXiv:2406.07089 [cs]

work page doi:10.48550/arxiv.2406.07089 2024
[46]

Jeasurk Yang, Sumin Lee, Sungwon Park, Minjun Lee, and Meeyoung Cha. 2024. https://doi.org/10.48550/arXiv.2410.09522 Poverty mapping in Mongolia with AI -based Ger detection reveals urban slums persist after the COVID -19 pandemic . arXiv preprint. ArXiv:2410.09522

work page doi:10.48550/arxiv.2410.09522 2024
[47]

Yi Yang and Shawn Newsam. 2010. https://doi.org/10.1145/1869790.1869829 Bag-of-visual-words and spatial extensions for land-use classification . In Proceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic Information Systems , GIS '10, pages 270--279, New York, NY, USA. Association for Computing Machinery

work page doi:10.1145/1869790.1869829 2010
[48]

Shuai Yuan, Guancong Lin, Lixian Zhang, Runmin Dong, Jinxiao Zhang, Shuang Chen, Juepeng Zheng, Jie Wang, and Haohuan Fu. 2024. Fusu: a multi-temporal-source land use change segmentation dataset for fine-grained urban semantic understanding. In Proceedings of the 38th International Conference on Neural Information Processing Systems, NIPS '24, Red Hook, N...

2024
[49]

online" 'onlinestring :=

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...
[50]

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 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...