On September 25, we (Tatjana Scheffler, Rafael Schirru, and Paul Lehmann) had the chance to present our work about data integration for points of interest from different community Web sites at the 35th German Conference on Artificial Intelligence in Saarbrücken.
POI matching has become an important task for different systems such as map applications and location recommendation systems. In recent years, users have contributed valuable information about locations (points of interest, POIs) in community projects such as OpenStreetMap (OSM) as well as in commercial social networks like Yelp or its German variant, Qype. These platforms often provide different types of information for the same objects, for example ratings (Qype), check-ins (Facebook Places), descriptions, categories, etc. For researchers and application developers it is often necessary to merge these distinct representations of POIs in order to obtain rich and complete information about the associated locations. Unfortunately the records representing the POIs do not share a common identifier across platforms thus making their matching a difficult task. In our work we present an approach matching POIs from Qype and Facebook Places to their counterparts in OSM. The algorithm uses different similarity measures taking the geographic distance of POIs into account as well as the string similarity of selected metadata fields. An evaluation of our system has shown that our proposed approach achieves an accuracy of 79% for data integration with Qype and 64% with Facebook Places while at the same time reducing the number of false matches compared to two baseline algorithms.
Details about the approach can either be found in our paper or can be directly obtained from the authors.