Recommendation of personalized social media content

When travelers plan their trips, landmark recommendation systems considering the properties of their trip will be convenient to help travelers determine locations they will visit.

We propose an approach to extracting and recommending landmark locations related to a user’s trip from social images. We first examine the impact of spatial and temporal properties of a trip on the distribution of popular places through large-scale data analysis. Second, we present an approach that efficiently extracts landmarks from social media and recommends user-centered landmarks. Our approach is to construct a vector that weights frequently visited places under the similar conditions of a traveler, and then to re-rank landmarks based on the vector. We provide users with the user-centered recommended landmarks, together with regions visited by travelers who visited the recommended landmarks together.


According to our experimental results, the approach enhanced the accuracy of recommended landmarks by 41% in the size of a city and by 65% in the size of a state, compared to a baseline. Through a user study, we also verified that it is applicable to lesser-known places where the social media images are small in quantity. Thus, we expect that the approach is helpful in developing personalized recommendations.