Automatic Tag Enrichment for Points-of-Interest in Open Street Map

Abstract

The user experience of geo-search engines and map services heavily depends on the quality of the underlying data. This is especially an issue for crowd-sourced data as e.g., collected and offered by the Open Street Map (OSM) project. In this paper we are focusing on points-of-interests (POIs), such as restaurants, shops, hotels and leisure facilities. Many of those are incompletely tagged in OSM (missing e.g., the amenity tag), which leads to such POIs not showing up in respective search queries or not being displayed correctly on the map. We develop methods that can automatically infer tags characterizing POIs solely based on the POI names. The idea being that many POI names already contain sufficient information for tagging. For example, ‘Pizzeria Bella Italia’ and ‘Chau’s Wok’ most certainly refer to restaurants, whereas ‘Cut & Color’ is likely a hairdresser. We employ machine learning techniques to extrapolate such additional tag information; our approach yields an accuracy of more than 85% for the considered tags. Moreover, for restaurants, we aimed for extrapolation of the respective cuisine tag (italian, sushi, etc.). For more than 19.000 out of 28.000 restaurants in Germany lacking the cuisine tag, our approach assigned a cuisine. In a random sample of those assignments 98% of these appeared to be true.