Semantically enriching personal mobility data using OpenStreetMap: a case study using smartphone users' frequently visited places.

Behavioral scientists use geodatabases to automatically annotate mobile device location data with semantically meaningful point-of-interest (POI) labels. Using volunteered geographic information (VGI), such as OpenStreetMap (OSM) data, to annotate large amounts of location data is more cost-effective and efficient than manual annotation by participants. The data quality of VGI, however, has been questioned, with limited evidence supporting its use for annotating personal mobility data. We assessed the performance of using OSM POI data to annotate year-long smartphone location data acquired from 93 people in the United States. A Python package was developed to extract POI geometric and semantic information from the OSM geodatabase and annotate places frequently visited by participants. We evaluated the semantic annotation performance of our OSM package against participant-provided annotations and compared OSM with two popular commercial geodatabases: Foursquare and Google Maps. Annotations acquired using OSM data had the best overall performance across eight categories of places, with 81% of places labeled and an average F1 score of 0.65, although Foursquare and Google Maps showed advantages for annotating some categories. This case study provides empirical evidence supporting the use of OSM for semantic enrichment in mobile device location data research. We outline recommendations for future implementations.