Figure 1.
A map of civil unrest event hotspots on September 27th, 2012 pertaining to labor reform and other issues.
Flags denote the ground-truth events reported by authorities. Circles denote the events detected by our method.
Figure 2.
Flowchart of the proposed method.
Table 1.
The algorithm of Dynamic Query Expansion.
Table 2.
The algorithm of Local Modularity Spatial Scan.
Table 3.
Dataset and Label Source.
Table 4.
Methods and Efficiencies.
Table 5.
Comparison between Expanded Query from DQE and GSR Description of Events.
Table 6.
Performance Comparison with Baseline Components (Precision, Recall, F-measure).
Table 7.
Performance Comparison with Existing Event Detection Methods (Precision, Recall, F-measure).
Figure 3.
Sensitivity analysis of parameters.
(a) Sensitivity analysis of “number of seed query terms” (b) Sensitivity analysis of “trade-off β for updating tweet node weights” (c) Sensitivity analysis of “trade-off between local modularity and spatial scan statistics”.
Figure 4.
Sensitivity analysis of the longest distance r between any two neighboring locations.
Figure 5.
Event detection case studies.