Automatic Generation of Temporal Feature Vectors with Application to Tourism Recommender Systems

Recommender systems have been widely used in our daily life to recommend objects to users meeting the users' preference. In this paper, we focus on objects with temporally variable features such as restaurant with seasonal dishes and point-of-interests (POIs) to have seasonal attractions, and propose a method to automatically generate temporal feature vectors for those objects. The basic idea of the proposed method is: 1) to identify the vocabulary concerned with objects through Wikipedia; 2) to identify the trend over all objects through Twitter; and 3) to highlight the weight of words contained in each identified trend to obtain temporal feature vectors for each object. We built a tourism recommender system to evaluate the effectiveness of the proposed method. The result of experiments indicates that: 1) the variance of temporal feature vectors follows the Gaussian distribution, 2) those vectors certainly reflect the similarity of POIs for a designated time period, and 3) such a property of feature vectors can be effectively used for the seasonal recommendation of POIs.