People search must map free-form location phrases to geographic entities used as structured retrieval filters. Lexical standardizers handle canonical names well but are brittle to aliases, misspellings, metropolitan expressions, and same-name ambiguity. We formulate this task as graded, set-valued entity retrieval over a fixed ontology. We identify three coupled design requirements: distinguishing identity-preserving variation from knowledge-dependent aliases, controlling false negatives among valid same-name entities, and separating stable transformations from mutable entity knowledge. We realize them in a prompt-asymmetric bi-encoder with calibrated alias support, bounded ambiguity-aware negatives, and editable entity documents that support localized updates without retraining. Across a fixed production-derived development benchmark and a public GeoNames transfer task, task adaptation improves substantially over frozen encoders and standard token baselines. Controlled development ablations show that specialized supervision contributes beyond standard task fine-tuning and encoder scaling. On GeoNames, the adapted model improves known-target Recall@1 throughout zero-to-moderate character overlap, while character n-grams retain a small aggregate Target Recall@5 advantage. In a blinded human comparison on a stratified production challenge set, our model raises relevant P@1 from 28.0% to 46.0% (p=0.012). Fixed-query endpoint estimates improve on non-canonical queries and remain close to control on frequent queries; a randomized live experiment detects no engagement regression. These results support task-adapted geographic entity retrieval as a practical replacement for the incumbent taxonomy-based standardizer, with the largest relevance gains on non-canonical queries.
Figures & tables
Figure 1. Prompt-asymmetric, shared-weight bi-encoder. Entity representations are precomputed; each query requires one encoding and an exhaustive filtered search. A shared bi-encoder precomputes geographic entity vectors offline and encodes each location query online before exact filtered retrieval.
Score
Interpretation
Example
4
Dominant referent
Los Angeles, California
3
Valid, non-dominant name match
La, Ghana
2
Geographically related, not denoted
Long Beach, California
1
Weak or incidental relation
New York, United States
0
No relation
Paris, France
Table 1. Geo-relevance rubric, illustrated for the query “la”.
Success@5
Model
≥3
≥4
nDCG@10
Open-source 0.6B
0.1240
0.0956
0.4581
Open-source 4B
0.2075
0.1548
0.6993
Open-source 8B
0.2816
0.2377
0.7124
Adapted 0.6B (ours)
0.8931
0.8499
0.9201
Table 2. Frozen open-source models versus task adaptation. Success@5 counts queries with at least one Top-5 result at or above the indicated grade threshold; bold marks the best result.
Configuration
Δ
nDCG@10
Cumulative supervision and ambiguity handling
Standard surface-form fine-tuning
—
0.8099
+ Initial ambiguity-aware sampling
+0.0511
0.8610
+ Refined ambiguity-aware sampling
+0.0087
0.8697
+ Validated query-side aliases
+0.0127
0.8824
+ Calibrated source sampling
+0.0067
0.8891
Table 3. Controlled ablations of supervision and entity representation. In the first block, each Δ is relative to the preceding row. In the second, JSON and validated-only compare with the NL reference; no-geohash compares with validated-only. Bold denotes the offline best; the dagger marks the deployed variant.
Measure
Before → after
Δ (95% CI)
Entity document
Alias Target R@1
0.182→0.644
+0.4616[+0.4574,+0.4658]
Alias Target R@5
0.287→0.766
+0.4795[+0.4753,+0.4837]
Canonical R@5
0.981→0.960
−0.0211[−0.0231,−0.0193]
Lexical alias table
Alias Target R@1
0.022→0.937
+0.9149[+0.9124,+0.9173]
Table 4. Incremental alias acquisition. Each mechanism is compared with its own pre-update state; Δ cells report entity-level paired-bootstrap 95% confidence intervals. Unshown disjoint controls are unchanged at four decimals.
Known source entity
Pooled graded relevance
Method
Target R@1
Target R@5
MRR@10
nDCG@10
P@1
Exact token
0.0155
0.0324
0.0227
0.1222
0.2570
Prefix token
0.0481
0.0805
0.0621
0.1111
0.2251
BM25
0.0622
0.1246
0.0891
0.1681
0.3137
Character n -gram
0.1064
0.2133
0.1525
0.2110
0.3550
Open-source 0.6B (frozen)
0.0424
0.0851
0.0607
0.1280
0.1356
Table 5. GeoNames alias transfer. Known-target metrics use each alias’s source entity; pooled metrics allow graded relevance to other valid entities. Bold marks the best result.
Slice
Pairs (share)
Char R@1
Ours R@1
Δ R@1 (95% CI)
Δ R@5 (95% CI)
J=0
236,608 (27.5%)
0.0000
0.0163
+0.0163[.0158,.0169]
+0.0408[.0399,.0417]
0<J≤0.25
202,469 (23.5%)
0.0101
0.0469
+0.0368[.0358,.0378]
+0.0784[.0769,.0799]
0.25<J≤0.50
241,499 (28.1%)
0.1177
0.1630
+0.0453[.0437,.0470]
−0.0055[−.0077,−.0033]
J>0.50
180,006 (20.9%)
0.3394
0.2642
−0.0752[−.0774,−.0730]
−0.1607[−.1632,−.1582]
Overall
860,582 (100%)
0.1064
0.1165
+0.0101[.0094,.0108]
−0.0055[−.0064,−.0046]
Table 6. Character-overlap analysis on GeoNames. Bold marks the higher Target Recall@1 per slice; ours leads for all J≤0.50 bands (79.1% of pairs) and overall, while character n -gram leads for J>0.50 . Deltas are ours minus character n -gram; 95% CIs use 10,000 query-clustered bootstrap replicates.
Metric
Prod.
Emb.
Comparison
Average grade@1 (0–4)
1.948
2.244
+0.296
Relevant P@1
28.0% (14/50)
46.0% (23/50)
+18.0 pp
Embedding W/T/L
—
—
26 / 14 / 10
Table 7. Blinded Top-1 comparison on the stratified production challenge set. W/T/L denotes embedding wins, ties, and losses.
Query set
Control
Geo Top-1
Geo Top-5
Traffic-derived frequent queries
2.7289
2.7390 ( +0.4% )
2.7223 ( −0.2% )
Non-canonical locations
3.3311
3.3997 ( +2.1% )
3.4959 ( +5.0% )
Table 9. Fixed-query endpoint relevance. Values are mean judged grade@10; Geo Top- k is geographic retrieval depth. Relative changes are computed from unrounded means. Descriptive point estimates only; no confidence intervals are available.
People search for urban outdoor places not only by category or function, but also by what activities a place can support and how it is perceived. Existing geospatial retrieval remains largely POIcentric and metadata-driven, making it difficult to satisfy openended, affective, or activity-oriented needs. We present PlaceSeek, a human-centered outdoor place retrieval framework that maps natural-language queries to geolocated street-view imagery. PlaceSeek introduces an intent-aware retrieval mechanism that decomposes user queries into functional and affective sub-intents. A Semantic Grounding Module verifies whether candidate street-view results contain the physical evidence needed to support the intended activity, while an Affective Alignment Module re-ranks physically valid candidates using a LoRA-adapted vision-language model trained on human urban perception judgments. We evaluate PlaceSeek on 31,956 street-view locations in Milan across 10 naturallanguage queries annotated by five human evaluators. PlaceSeek achieves 88.0% Precision@5, a mean match score of 3.39/4.0, and 0.920 nDCG@5, outperforming CLIP, fine-tuned CLIP, SigLIP, and a VQA-based baseline. Ablation results show that physical grounding is essential for retrieval validity, while affective alignment improves ranking quality among physically valid candidates. These findings highlight that complex urban spatial queries require modeling both verifiable visual evidence and human perceptual preferences. PlaceSeek provides a potential framework for human-centered nextgeneration geospatial retrieval systems.
Ziqi Cui, Shangyu Lou
University of California, Santa Barbara & San Diego State University California, USA
This paper addresses automatic geospatial question answering over multilingual toponymic data. An original bilingual dataset of toponyms of the Republic of Tatarstan is introduced, comprising 9,688 structured records with linguistic, etymological, administrative, and coordinate information (93.1% georeferenced). Based on this dataset, a question-answering corpus of approximately 39,000 question-context-answer triples is constructed with guaranteed answer localization. A hybrid retriever integrates dense semantic indexing (multilingual-e5-large) with geospatial filtering via KD-trees and haversine distance. On 500 test queries, the hybrid search achieves Recall@1=0.988, Recall@5=1.000, and MRR=0.994, significantly outperforming BM25 and purely spatial methods. Among tested reader architectures (RuBERT, XLM-RoBERTa-large, T5-RUS), XLM-RoBERTa-large attains the best quality: EM=0.992, F1=0.994. On raw outputs, RuBERT models fail on coordinate questions (F1=0) while XLM-RoBERTa-large reaches F1=0.984; however, simple post-processing eliminates numerical gaps and restores RuBERT accuracy to 100%. This discrepancy stems from tokenization differences and pre-training corpora composition. All resources (dataset, QA corpus, model weights, web demo) are openly published on Hugging Face. Results apply to geospatial QA services, geocoding, and digital humanities in multilingual regions.
Mullosharaf K. Arabov
Department of Data Analysis and Programming Technologies Institute of Computational Mathematics and Information Technologies Kazan (Volga Region) Federal University Kazan, Russia
Company name retrieval and final entity clustering can respond differently to encoder adaptation. We study this dependency through a controlled pipeline: train multilingual E5 encoders with matched positive examples and random or hard negatives, select checkpoints on development data, retrieve within equal country-specific edge budgets, and apply unchanged pair classifiers and guarded Union-Find clus- tering. Evaluation uses a subsequently collected GLEIF snapshot; a stratified, assisted human audit and a post-hoc temperature sweep examine label and parameter sensitivity. On 3,633 names with source-silver labels, adapted retrieval raises direct pair recall from 53.74% to 76.98%, whereas cluster recall moves from 32.54% to 33.33%. Most newly retrieved positives fail the existing classifier threshold. The study contributes a controlled comparison and pair-level diagnosis of this loss between retrieval and confirma-tion. It shows why evaluating a replacement retriever requires measuring the resulting entity partition, while leaving joint matcher retraining and independent gold-label evaluation for future work.
Yijiashun Qi, Yuxuan Li, Hanzhe Guo
University of Michigan, Ann Arbor, USA · University of Pennsylvania, Philadelphia, PA, USA