Towards rule-guided classification for volunteered geographic information

Abstract

Crowd-sourcing, especially in form of Volunteered Geographic Information (VGI) significantly changed the way geographic data is collected and the products that are generated from them. In VGI projects, contributors' heterogeneity fosters rich data sources, however with problematic quality. In this paper, we tackle data quality from a classification perspective. Particularly in VGI, data classification presents some challenges: In some cases, the classification of entities depends on individual conceptualization about the environment. Whereas in other cases, a geographic feature itself might have ambiguous characteristics. These problems lead to inconsistent and inappropriate classifications. To face these challenges, we propose a guided classification approach. The approach employs data mining algorithms to develop a classifier, through investigating the geographic characteristics of target feature classes. The developed classifier acts to distinguish between related classes like forest, meadow and park. Then, the classifier could be used to guide the contributors during the classification process. The findings of an empirical study illustrate that the developed classifier correctly predict some classes. However, it still has a limited accuracy with other related classes.