SocialMix: A familiarity-based and preference-aware location suggestion approach

Abstract

Traditionally, location suggestion systems have employed collaborative filtering model to make recommendations for users based on data gathered from users with similar interests, demographics, and check-in records. However, these techniques fail to take into account on very important element present in online social networks, the online relationships that these users maintain. Arguably, this is the most important aspect of their online profiles, often more revealing than their self reported personal interests and check-in records. Aiming to improve the accuracy and novelty of recommendations, this research proposes a hybrid location suggestion algorithm, called SocialMix, of which, takes into full consideration a user’s familiarity and preference (interest) similarity, along with relationships. In the first part of this study, we compute the degrees of familiarity between users using three feature variables: the number of mutual friends, the Jaccard index and cosine similarity. In order to determine the weights of the these feature variables, maximum likelihood estimation used, and then the features are fit to a Logistic Regression model in order to calculate the degrees of familiarity. The second part of this research we present a new method for calculating similarity between individuals by integrating users familiarity and preference similarity. This allows us to introduce a new location interest degree calculation method on the hybrid similarity. Extensive experiments were conducted on several real datasets. The performance of SocialMix was analyzed for both accuracy and time complexity using the following metrics: MAE (mean absolute error), RMSE (root mean square error), Precision, Recall, F-measure, Coverage rate, Popularity and Response time. Results were compared against classical recommendation approaches as a baseline. The results show that the accuracy and time performance of SocialMix, when compared with other algorithms which do not consider social relationships, are demonstratively improved. In addition, a positive by product worth noting is that SocialMix has a tendency to recommend more obscure but still interesting locations.