Understanding Tourist Destination Choices from Geo-tagged Tweets

Abstract

Tourism related travels have significant impacts on transportation infrastructures, especially in large tourist attractions such as Florida. It is very expensive to collect individual travel data of a reasonable number of tourists traveling over a large region. Ubiquitous use of social media allows us to collect tourist travel data at a large scale in a cost effective way. This paper presents an analysis of tourist destination choices with longitudinal travel data collected from Twitter. From a collection of geo-tagged tweets, we have filtered out a reliable sample and identified tourists using a data mining approach. Then we find the tourists' destinations inside Florida. We have created a sequence of visited locations and applied a Conditional Random Field (CRF) model to predict the type of a tourists' next destination. The proposed model utilizes the features extracted from tweet posted time and location types. The feature set can be expanded by incorporating content-based features without violating the assumptions of CRF. The data collection steps and results derived from this study will be significantly useful for building an individual-level travel behavior model for tourists using social media data.