Learning to Predict Tourists Movements
- Day - Time: 03 June 2013, h.11:00
- Place: Area della Ricerca CNR di Pisa - Room: C-29
Touristic applications stirred an increased interest in the last years due to the intense use of mobile devices and location based applications. Our outlook on this matter is directed towards the next point of interest (PoI) prediction task. We tackle the problem of predicting the â??nextâ?? geographical position of a tourist, given her history (i.e., the prediction is done accordingly to the touristâ??s current trail) by means of supervised learning techniques.
We test our methods on three datasets built using geo-tagged pictures downloaded from Flickr, each collection corresponding to a popular touristic area. We adopt two popular Machine Learning methods, namely Gradient Boosted Regression Trees and Ranking SVM for learning to rank the next PoI, on the basis of an object space represented by a multi-dimensional feature vector, specifically designed for tourism related data. We define a set of 68 different features, broadly classified into two main categories, namely â??Sessionâ?? and â??PoIâ??. Session features are meant to model the tourist behavior and capture concepts like groups of PoIs visited, distances among PoIs and other characteristics of a user session (trail). On the other hand, PoI features model the characteristics of a candidate PoI, also taking into account the past activities of the tourist.
We propose a thorough comparison of several methods that are considered state-of-the-art in touristic recommender and trail prediction systems (WhereNext, Random Walk with Restart), as well as a strong popularity baseline. As experiments show, the methods we propose constantly outperform, with up to 300% in terms of prediction accuracy, our baselines and provide strong evidence of the performance and robustness of our solutions.
NOTE: This seminar is the fifth one of the series of six seminars presented by the winners of the prize "Young researchers ISTI 2013". Cristina Muntean placed third in the PhD student category.