Improving an Open Source Geocoding Server

Abstract: A common problem in geocoding is that the postal addresses as requested by the user differ from the addresses as described in the database. The online, open source geocoder called Nominatim is one of the most used geocoders nowadays. However, this geocoder lacks the interactivity that most of the online geocoders already offer. The Nominatim geocoder provides no feedback to the user while typing addresses. Also, the geocoder cannot deal with any misspelling errors introduced by the user in the requested address. This thesis is about extending the functionality of the Nominatim geocoder to provide fuzzy search and autocomplete features. In this work I propose a new index and search strategy for the OpenStreetMap reference dataset. Also, I extend the search algorithm to geocode new address types such as street intersections. Both the original Nominatim geocoder and the proposed solution are compared using metrics such as the precision of the results, match rate and keystrokes saved by the autocomplete feature. The test addresses used in this work are a subset selected among the Swedish addresses available in the OpenStreetMap data set. The results show that the proposed geocoder performs better when compared to the original Nominatim geocoder. In the proposed geocoder, the users get address suggestions as they type, adding interactivity to the original geocoder. Also, the proposed geocoder is able to find the right address in the presence of errors in the user query with a match rate of 98%.