LLM-Driven Training-free Location-Attribute Synergic Fusion: A Closed-Loop Paradigm for Dual-source Encrypted POIs and LULC Mapping

Chang Li; Xingtao Peng; Yongjun Zhang; Yinfei He; Cairun Huang

cs.CLSep 6, 2026

LLM-Driven Training-free Location-Attribute Synergic Fusion: A Closed-Loop Paradigm for Dual-source Encrypted POIs and LULC Mapping

Authors: Chang Li, Xingtao Peng, Yongjun Zhang, Yinfei He, Cairun Huang

Organizations: Key Laboratory for Geographical Process Analysis & Simulation of Hubei Province, and College of Urban and Environmental Sciences, Central China Normal University, Wuhan, 430000, HuBei, China · School of Remote Sensing and Information Engineering, Wuhan University, Wuhan, 430000, HuBei, China

Abstract

Dual-source encrypted points of interest (DSEP), POIs from two encrypted coordinate systems, suffer from intertwined location and attribute uncertainties, including nonlinear systematic misalignment and naming inconsistency, hindering land-use/land-cover (LULC) mapping. To the best of our knowledge, this paper is the first to propose an LLM-driven, training-free location-attribute synergic closed-loop optimization paradigm for DSEP fusion. The paradigm jointly refines location transformation and attribute correspondences through iterative feedback. Attribute-synergic location fusion uses an LLM-driven attribute matching method to establish DSEP correspondences, reducing matching complexity from O(N^2) to O(N), and refines transformation coefficients using an improved particle swarm optimization algorithm within ISODATA-clustered local subregions. Location-synergic attribute fusion then reassesses attribute confidence from updated geometric residuals through an LLM-fuzzy method. The refined correspondences feed back into location optimization, forming a bidirectional closed loop. Sample purification and adaptive radius contraction enable convergence in essentially two iterations. We further propose a training-free LULC mapping method that inherits land-use classes from encrypted maps through location fusion, producing vector-raster integrated LULC maps. A reference-free POI fusion evaluation method is applied across 31 provincial capitals and municipalities in mainland China. Experiments show that our method achieves an average DSEP location fusion residual of 4.58 m and attribute fusion accuracy of 95.12%, improving upon the open-source baseline and state-of-the-art method by 1.77 m and 14.87%, respectively. Overall, the method provides a training-free solution for DSEP fusion and enables georeferencing of encrypted vector data to WGS-84 without field-surveyed ground control points.

Figures & tables

Figure 1: Illustration of coordinate offset and attribute inconsistency in dual-source encrypted points of interest (DSEP) fusion. Although POI attributes or locations are different in multiple sources, location and attribute are interdependent for the same geographic entity and can synergically support each other in the DSEP fusion.

Attribute of Location	Baidu Maps	Amap
Name	北京工业大学附属中学(双桥分校)	北京工大附中双桥分校
English Name	Beijing University of Technology Affiliated High School (Shuangqiao Branch)	Affiliated High School of BJUT, Shuangqiao Branch
Address	双桥中路32号院	双桥中路32号院
English Address	No. 32 Courtyard, Shuangqiao Middle Road	No. 32 Courtyard, Shuangqiao Middle Road
Coordinate (longitude, latitude)	116.596075, 39.899951	116.589449, 39.894216

Table 1: Semantic differences in the representation of the same POI between Baidu Maps and Amap.

Method category	Attribute matching	Coordinate registration	Uncertainty treatment	Interdependence
String matching + global transform [ 34 , 5 , 17 ]	Edit distance / Jaccard	Global OLS / affine	None	Independent, sequential
Deep learning matching [ 11 , 28 , 20 ]	BERT / contrastive	Global / local transform	Implicit (during training)	Independent, sequential
Graph-based matching [ 16 ]	Graph node matching	Not separately handled	None	Independent
Multi-source prediction [ 30 ]	Not handled	GNN prediction	Learnable fidelity	Independent
This paper	LLM / any model	RANSAC + PSO local optimization	Explicit, bidirectional	Interdependent, synergic

Table 2: Comparison with existing POI fusion approaches across key differential axes.

Figure 2: Research route. The steps of the algorithm are as follows: I.Global POI fusion begins by aligning textual attributes and coordinates to obtain an initial set of fused data. II. Local POI fusion is conducted by dividing the data into spatial subregions. Within each subregion, POI attributes are preliminarily fused using an LLM. III.The intelligent optimization algorithm further reduces location residuals. IV. Fuzzy mathematics is employed to resolve attribute ambiguities, resulting in the final fused POI dataset. The dashed loop between Attribute-synergic location fusion and Location-synergic attribute fusion indicates the iterative loop: attribute correspondences guide location refinement, and the refined geometry in turn re-conditions attribute confidence, forming a bidirectional feedback that progressively co-optimizes both stages until convergence.

Figure 3: The local POI matching algorithm based on the local search circle

Figure 4: The basic architecture of DeepSeek-Chat [ 7 ] , which is similar to the transformer block and consists of multiple blocks composed of multi-head latent attention (MLA), a DeepSeekMoE-based feedforward network, and root mean square layer normalization (RMSNorm).

Type of prompts	Content of prompts
Environment Prompts	In subsequent interactions, I will provide two formatted Chinese addresses, separated by an underscore. You are required to determine purely based on semantics whether the two addresses refer to the same location and respond with a confidence score between 0 and 1 (where 0 indicates they are definitely not the same address and 1 indicates they are definitely the same address).
Judgment Rules	Determine whether the addresses refer to the same location based on the following principles: If a larger-scale location contains a smaller-scale location, they should be considered different locations. For instance, "Beijing International Studies University" and "Beijing International Studies University-Qinxue Building" should be identified as different locations. If the descriptions differ but the addresses are semantically identical, they should be recognized as the same address. For example, "No. 1, Lane 3, Dong’an Street" and "No. one, Third Lane, Dong’an Street, Fengtai Town" should be considered the same address. Please note that I will provide multiple groups of addresses (different groups will be separated by "\|"). Please respond in the same sequence and format accordingly.
Manually set the content of the dialog 1	Q : "Beijing International Studies University""Beijing International Studies University"\|"Building 307, Yard 28, Guangqu Road""Building 307, Yard 28, Guangqu Road"
	A : 0.98\|0.95
Manually set the content of the dialog 2	Q : "Beijing International Studies University""Beijing Foreign Studies University"\|"No. 2, Wenxing Street""Near No. 2, Wenxing Street"
	A : 0.02\|0.93

Table 3: Predefined prompts. By incorporating environment prompts, judgment rules, and predefined multi-round dialogues, the LLM produces structured outputs, enabling efficient aggregation of multi-source POI attribute matches.

Similarity	Levels	Similarity	Levels
$1>s_{A,ij}\geq 0.9$	I	$0.9>s_{A,ij}\geq 0.8$	II
$0.8>s_{A,ij}\geq 0.6$	III	$0.6>s_{A,ij}\geq 0.4$	IV
$0.4>s_{A,ij}\geq 0.1$	V	$0.1>s_{A,ij}\geq 0$	VI

Table 4: Uncertainty levels of POIs attribute fusion. Levels I: Exactly the same, II: Very similar, III: Basically similar, IV: Uncertain, V: Dissimilar, and VI: Completely inconsistent.

Figure 5: Study areas. The basemap is based on approval No. GS (2024) 0650 reprojected into an equidistant cylindrical projection to ensure consistent latitude and longitude intervals.

Figure 6: Illustration of residual coordinate (geometric) errors in X/Y (longitude and latitude) direction. Due to space limitations, subfigures (a), (b), (c), and (d) show residuals for representative cities located in the north, east, south, and west of China (Harbin, Shanghai, Haikou, and Urumqi), respectively, where the randomness of residuals can be found in different directions.

LLMs	Algorithm	Performance Metrics (m)
LLMs	Algorithm	Mean	Median
	Origin	1286.74	1286.95
ChatGPT	FSR-PSO	888.67	941.54
	$\mathrm{PSO_{Poly}}$	598.05	577.31
	$\mathrm{PSO_{Trig}\text{-}OLS}$	904.68	964.64
	GOLS	915.28	978.8

Table 5: Location registration error of particle swarm optimization (PSO) based on different objective functions.

\mathrm{PSO_{Poly}}

represents the objective function based on formula 13 , while FSR-PSO refers to the process of applying

\mathrm{PSO_{Poly}}

to the coordinates followed by OLS. The same applies to

\mathrm{PSO_{Trig}\text{-}OLS}

based on the objective function formula 15 .

Figure 7: Illustration of residual coordinate (geometric) errors in X/Y (longitude and latitude) direction after global-to-local location fusion (GTLLF) .

Method	Mean (m)	Std (m)	Median (m)
Origin	1286.74	2.94	1286.68
TPS	17.65	9.87	15.58
ETT	6.35	3.92	5.49
GTLLF (Ours)	4.58	2.82	3.91

Table 6: The comparison of residuals for different methods. Statistics were aggregated from per-city results. Detailed results for all cities are provided in Table S2.

Figure 8: Sensitivity analysis of hyperparameters on fusion residual Mean. (a) iterative self-organizing data analysis (ISODATA) $k_{\text{init}}$ , (b) ISODATA split threshold, (c) particle swarm optimization (PSO) cognitive coefficient $c_{1}$ , (d) PSO social coefficient $c_{2}$ , (e) PSO inertia weight $\omega$ .

LLM	Method	Average (m)	Median (m)
ChatGPT	Origin	1723.24	1723.01
	ETT	37.55	36.47
	TPS	293.98	297.13
	GTLLF (Iter-1)	54.19	58.41
	GTLLF (Iter-2)	37.36	36.29
	GTLLF (Iter-3, Final)	37.36	36.29

Table 7: Ground truth validation of fusion accuracy using the synthetic encrypted dataset. For GTLLF, Iter-1, Iter-2 and Iter-3 correspond to the residuals after the first, second and third outer synergic iterations respectively, where Iter-3 is the reported final result. Bold values indicate the best performance among empirical trigonometric transformation (ETT), thin-plate spline (TPS) and global-to-local location fusion (GTLLF) for each LLM.

Figure 9: Attribute fusion evaluation results

Figure 10: Google Maps LULC data in Kunming, China. Within mainland China, Google Maps provides only basic categories and lacks fine-grained urban LULC elements for some cities, and its data updates are infrequent, limiting its applicability as a LULC reference.

Method ( $N$ )	mIoU (%)	Macro-F1 (%)
LEPF-LULC (Ours, $N$ =31)	75.14 $\pm$ 5.74	83.11 $\pm$ 4.90
EULUC ( $N$ =26)	58.58 $\pm$ 6.83	72.01 $\pm$ 5.74

Table 8: Quantitative comparison of land-use/land-cover (LULC) classification accuracy between EULUC and the proposed joint LLM- and encrypted POI-driven training-free LULC mapping (LEPF-LULC). Metrics are mean Intersection-over-Union (mIoU) and Macro-F1, reported as mean

\pm

standard deviation (%) across cities.

N

denotes the number of cities evaluated. EULUC results were computed over 26 cities where EULUC provides coverage; cities with no EULUC data were excluded from the mean calculation for fairness. Detailed results for all cities are provided in Table S3.

City	Direct Matching (times)	Fixed Distance Matching (times)	Local Search Circle (times)
Shanghai	46958	24.83	1.35 (94.56% $\downarrow$ )
Shenzhen	34069	32.58	1.34 (95.89% $\downarrow$ )
Guangzhou	34758	24.00	1.45 (93.96% $\downarrow$ )
Average	36202	27.14	1.38 (94.92% $\downarrow$ )

Table 9: Comparison of points of interest (POI) pairwise computations. The proposed local search circle significantly reduces computations, demonstrating much higher efficiency compared to exhaustive direct matching and fixed distance matching with a predefined search radius.

Remote Sensing

Remote sensing uses satellite, aerial, and ground-based observations to measure and map Earth and other planetary surfaces.

See all 528 papers

LoCA: Enabling Multi-Modality in Geospatial Location Encoders
Oct 23, 2026Panagiotis Koromilas, Eleanna Vali, Diogenis Antonopoulos +7Geospatial Representation Learning Multimodal Contrastive Learning
University of Athens · The Cyprus Institute · Archimedes, Athena Research Center +7
CartoVec: General-Purpose Location Embeddings from Open Map Data
Oct 12, 2026Furkan GursoyGeospatial Representation Learning Spatial Context Modeling
3D LULC classification using multispectral LiDAR and deep learning: current and prospective schemes
May 21, 2026Narges Takhtkeshha, Aldino Rizaldy, Markus Hollaus +3Remote Sensing Image Understanding Point Cloud Classification
3D Optical Metrology (3DOM) Unit, Bruno Kessler Foundation (FBK), Trento, Italy · Department of Geodesy and Geoinformation, TU Wien, Vienna, Austria · Helmholtz-ZentrumDresden-Rossendorf (HZDR), Helmholtz Institute Freiberg for Resource Technology (HIF), Freiberg, Germany +2

Oct 23, 2026J/K move · Enter open · S save

LoCA: Enabling Multi-Modality in Geospatial Location Encoders

Panagiotis Koromilas, Eleanna Vali, Diogenis Antonopoulos, Eleni Saka+6

University of Athens · The Cyprus Institute · Archimedes, Athena Research Center+7

Geospatial Representation Learning Multimodal Contrastive Learning Remote Sensing

Open paper AskPDF arXiv

Location encoders map geographic coordinates to high-dimensional descriptors through contrastive alignment with paired Earth Observation imagery. We prove that this construction is structurally limited: any Lipschitz location encoder produces an output of intrinsic dimension at most 2, regardless of architecture or bandwidth, which forces the contrastive loss to strictly exceed its asymptotic optimum. Empirically, joint training also collapses the encoder's embedding covariance to low rank. Although dimension-limited, current location encoders sidestep this spectral collapse by training against a frozen image encoder, which we reinterpret as a covariance regularizer that supplies a high-rank target for the location encoder to fold onto. However, extending this approach to multimodal inputs is challenging, as location encoders should simultaneously map a single location to multiple, potentially incompatible, descriptors. We identify two requirements a regularizer must satisfy for location encoding in multimodal regimes: co-alignment across modalities, since adapters alone cannot align independently pretrained encoders; and high effective rank, since the location encoder inherits its target's spectrum. We propose LoCA (Location Contrastive Alignment), a two-stage pipeline that satisfies both. Stage 1 jointly trains Sentinel-1 and Sentinel-2 encoders with contrastive learning, producing a co-aligned, high-rank multimodal target. Stage 2 freezes the image encoders and trains a location and a tabular metadata encoder as projection adapters with multi-view contrastive learning. Against five state-of-the-art encoders on 17 tasks ranging from land cover and wildfire forecasting to geophysical, socio-economic, biodiversity and climate applications, LoCA achieves the most task wins, is the only model to outperform every other on a majority of tasks, and wins 68.8% of all per-task comparisons.

Abstract

Remote Sensing

Explore similar work

LoCA: Enabling Multi-Modality in Geospatial Location Encoders

CartoVec: General-Purpose Location Embeddings from Open Map Data

3D LULC classification using multispectral LiDAR and deep learning: current and prospective schemes