notesum.ai

Published at November 9

Multimodal Contrastive Learning of Urban Space Representations from POI Data

cs.AI

Released Date: November 9, 2024

Authors: Xinglei Wang¹, Tao Cheng¹, Stephen Law², Zichao Zeng³, Lu Yin, Junyuan Liu¹

Aff.: ¹SpaceTimeLab, Department of Civil, Environmental and Geomatic Engineering, University College London, London, WC1E 6BT, United Kingdom; ²Department of Geography, University College London, London, WC1E 6BT, United Kingdom; ³3DIMPact, Department of Civil, Environmental and Geomatic Engineering, University College London, London, WC1E 6BT, United Kingdom

Arxiv: http://arxiv.org/abs/2411.06229v1

Model	Linear			MLP
Model	Precision ↑	Recall ↑	F1 Score ↑	Precision ↑	Recall ↑	F1 Score ↑
Random	9.6 ± 0.7	10.3 ± 0.5	9.7 ± 0.5	8.8 ± 1.3	10.3 ± 0.3	9.0 ± 0.3
TF-IDF	31.5± 0.4	32.2 ± 0.2	31.3 ± 0.3	31.8 ± 0.6	33.3 ± 0.5	31.7 ± 0.6
LDA	30.8 ± 0.3	29.1 ± 0.2	28.4 ± 0.2	31.5 ± 1.1	30.4 ± 0.7	29.2 ± 0.9
Place2Vec	30.9 ± 0.8	26.1 ± 0.7	26.3 ± 0.7	35.1 ± 1.2	32.7 ± 1.0	32.4 ± 1.2
Doc2Vec	32.4 ± 0.4	28.2 ± 0.1	28.0 ± 0.1	34.9 ± 0.9	33.8 ± 0.5	32.7 ± 0.6
SPPE	30.5 ± 0.4	27.0 ± 0.2	26.6 ± 0.2	34.5 ± 0.9	32.9 ± 0.7	32.2 ± 0.5
HGI	33.0 ± 0.5	30.0 ± 0.6	29.9 ± 0.6	33.6 ± 0.5	32.0 ± 0.9	31.6 ± 0.7
Space2Vec	28.6 ± 0.6	28.5 ± 0.8	27.4 ± 0.7	29.6 ± 0.6	28.9 ± 0.52	27.8 ± 0.3
CaLLiPer-SenTrans	$\mathbf{37.2\pm 0.4}$	$\mathbf{36.5\pm 0.6}$	$\mathbf{36.0\pm 0.3}$	$\underline{38.4\pm 1.5}$	$\mathbf{36.8\pm 1.5}$	$\mathbf{36.4\pm 1.6}$
CaLLiPer-Llama	$\underline{34.5\pm 1.0}$	$\underline{34.6\pm 1.0}$	$\underline{33.8\pm 0.9}$	$\mathbf{38.8\pm 1.0}$	$\underline{35.8\pm 0.85}$	$\underline{35.5\pm 1.0}$