Graph-Structured Trajectory Extraction from Travelogues

TL;DR

This study proposes a graph representation that retains information on the geographic hierarchy as well as the temporal order of visited locations, and has constructed a benchmark dataset for graph-structured trajectory extraction.

Abstract

Previous studies on sequence-based extraction of human movement trajectories have an issue of inadequate trajectory representation. Specifically, a pair of locations may not be lined up in a sequence especially when one location includes the other geographically. In this study, we propose a graph representation that retains information on the geographic hierarchy as well as the temporal order of visited locations, and have constructed a benchmark dataset for graph-structured trajectory extraction. The experiments with our baselines have demonstrated that it is possible to accurately predict visited locations and the order among them, but it remains a challenge to predict the hierarchical relations.

Figures (5)

• Figure 1: Illustration of our proposed tasks: visit status prediction (VSP) and visiting order prediction (VOP). The goal of VSP is to assign visit status labels to mentions (top) and to entities (middle). The goal of VOP is to build a visiting order graph by identifying inclusion and transition relations between entity pairs (bottom).
• Figure 2: Example of a visiting order graph, the same example at the bottom of Figure \ref{['fig:step4_annotation_overview']}.
• Figure 3: Illustrations of the gold visiting order graph and LUKE's prediction for an actual document (ID 00019). The nodes with dashed frames and the edges with dashed arrows represent prediction errors.
• Figure 4: Confusion matrix of LUKE for mention-level visit status prediction.
• Figure 5: Confusion matrix of LUKE for entity-level visit status prediction.