Geo-Link: Correlations of Heterogeneous Geo-Spatial Entities

Abstract

Correlations between geo-spatial entities in the real world are universal, and the proper exploitation of such correlations can aid in the understanding and analyzing of geo-spatial data. To improve geo-spatial data retrieval performance, we address the geo-link problem of heterogeneous geo-spatial entities. First, a graph-like multi-view correlation model is proposed that uses feature links to obtain information regarding entities and their interconnections. The proposed model construction algorithm can simultaneously incorporate geo-location, semantic description, visual content and their mutual correlations. It exploits the underlying relationship between the features and calculates the strength of the correlation between the different entities. An improved block-based strategy is then employed that takes all of the different features into consideration, where different features representing various semantic meanings are weighted according to neighboring entities. This allows for an expedited construction process. Experimental results using real data support the superiority of this approach in terms of both speed and comprehensiveness when compared to traditionally used methods.