A personalized news recommendation using user location and news contents

Personalized news article recommendation provides interesting articles to a specific user based on the preference of the user. With increasing use of hand-held devices, the interests of users are not influenced only by news contents, but also by their location. Therefore, in this paper, we propose a novel topic model to incorporate user location into a user preference for the location-based personalized news recommendation. The proposed model is Spatial Topical Preference Model (STPM). By representing the preference of a user differently according to the location of the user, the model recommends the user appropriate news articles to the user location. For this purpose, we represent geographical topic patterns with a Gaussian distribution. STPM is trained only with the news articles that the user actually reads. As a result, it shows poor performance, when the user reads just a few news articles. This problem of STPM is compensated for by LDA-based user profile that is not affected by user location. Therefore, the final proposed model is a combined model of STPM and LDA. In the evaluation of the proposed model, it is shown that STPM reflects user locations into news article recommendation well, and the combined model outperforms both STPM and LDA. These experimental results prove that the location-based user preference improves the performance of news article recommendation, and the proposed model incorporates the locational information of users into news recommendation effectively.