University of Amsterdam at the TREC 2013 Contextual Suggestion Track: Learning User Preferences from Wikitravel Categories

This paper describes our participation in the TREC 2013 Contextual Suggestion Track. The goal of the track is to evaluate systems that provide suggestions for activities to users in a specific location, taking into account their personal preferences. As a source for travel suggestions we use Wikitravel, which is a community-based travel guide for destinations all over the world. From pages dedicated to cities in the US we extract suggestions for sightseeing, shopping, eating and drinking. Descriptions from positive examples in the user profiles are used as queries to rank all suggestions in the US. Our user-dependent approach merges the per-query rankings of the positive examples of a single user. We automatically classified the rated examples according to the Wikitravel categories—Buy, Do, Drink, Eat and See— and derived a user-specific prior probability per category. With these we re-rank Wikitravel suggestions. The ranked suggestions are then filtered based on the location of the user.