Multimodal spatial co-occurrence knowledge representation for virtual trajectory classification

doi:10.1080/17538947.2026.2738310

Multimodal spatial co-occurrence knowledge representation for virtual trajectory classification

2026-09-28

0

OA

AI

DOI：10.1080/17538947.2026.2738310

Original

Original request for help

Save

Abstract

En 中文

Virtual trajectories, defined as temporally ordered records of user interactions with public map service platforms, provide valuable evidence of domain-oriented map use. Classifying virtual trajectories across application domains is important for understanding heterogeneous map-use purposes and supporting personalised geospatial services. Existing methods are predominantly POI-centred, and therefore provide limited representation of the broader map environment perceived by users during browsing. This study proposes a virtual trajectory classification framework based on a multimodal spatial co-occurrence knowledge graph (MSCKG). We developed an interpretable composite semantic taxonomy for map tiles to characterise the functional composition of tile contexts and then constructed an MSCKG integrating POIs, tiles, and category attributes through explicit spatial relations. Knowledge graph embedding was employed to learn unified representations of heterogeneous spatial elements, which were subsequently aggregated for domain classification. The bilinear knowledge graph embedding model DistMult achieved the best performance in experiments using Tianditu access logs from six application domains in Beijing, with an accuracy of 0.756, outperforming the baseline by 0.320. The proposed tile taxonomy identified 117 valid composite categories and provided interpretable semantic support for representing tile contexts. Data-intrinsic spatial indicators further revealed how tile semantics, POI contexts, and domain-specific spatial organisation jointly shape classification capacity.

Keywords:

Virtual trajectories

virtual trajectory classification

tile taxonomy

multimodal spatial co-occurrence knowledge graph

public map service platforms

Multimodal spatial co-occurrence knowledge representation for virtual trajectory classification

Abstract

Journal

Organization

Cited Papers