How to Extract Seasonal Features of Sightseeing Spots from Twitter and Wikipedia (Preliminary Version)

Abstract

In this paper, we consider a tourism recommender system which can recommend sightseeing spots for users who wish to make a travel plan for a designated time period such as early autumn and Christmas vacation. A key issue in realizing such seasonal recommendations is how to calculate feature vector of each spot which would vary depending on the time of travel. We propose a two-phase scheme which generates seasonal feature vectors for each sightseeing spot. In the first phase, the basic feature vector is generated for each spot using the description of Wikipedia and the TF-IDF weights. In the second phase, seasonal feature vectors are generated for each spot by referring to the distribution of keywords contained in tweets associated with spots for each season. The performance of the scheme is evaluated via experiments using actual data set drawn from Wikipedia and Twitter.