Recommendation of Shopping Places based on Social and Geographical Influences

The project tackled in this article is a shopping recommender system that aims at providing recommendations of new interesting shopping places to users, by considering their tastes and those of their friends, since social friends are often sharing common interests. This kind of system is a Location-Based Social Network. It considers social relationships and check-ins; i.e. the action of visiting a shopping place. In order to recommend shopping places, we are proposing a method combining three separated graphs, namely the social graph, the frequentation graph and a geographic graph into one graph. Hence, in this merged graph, nodes can represent users or places, and edges can connect users to each other (social links), users with places (frequentation relations) or places to each other (geographic relations). Given that check-in behavior of users is strongly dependent on the distances, the geographic graph is constructed considering the density of probabilities that a check-in is done according to its distance to the other check-ins. The Katz centrality is then used on the merged graph to compute the scores of candidate locations to be recommended. Finally, the top-n unvisited shopping places are recommended to the target user. The proposed method is compared to methods from the literature on a real-world datatset. The results confirm the real interest of considering both social and geographic data beyond the frequentations for recommending new places. Generally, our method outperforms significantly the compared methods, but under certain conditions that we analyze, we show it gives sometimes mixed results.