Recommending OSM Tags To Improve Metadata Quality

Abstract

In this paper an application is developed that functions similar to a recommender system and allows to find appropriate OpenStreetMap (OSM) tags by querying co-occurring keys and tags, as well as similar sets of tags in the database. A user may enter key(s) or key-value pair(s), even using wildcard substitution for both, in order to find keys or key-value pairs that are used in combination with the entered ones. Moreover, the top-k matching tag sets are also presented. The results are then top-k ranked, based on the frequency of the occurrence of each distinct set in the database. This information may enable a user to find the most comprehensive and best fitting tag set for an OSM element. This assumption is examined in an evaluation where the precision and recall metrics for both approaches are computed and compared. Our approach helps discovering combinations of tags and their usage frequency in contrast to common recommender systems that focus on classifying or clustering elements and finding the most accurate (single) class or cluster rather than sets of tags.