Social networks contain information regarding connections amongst members, responses of members to various posts, profile information, and members’ preferences for certain geographical sites or locations. Analysis of social networks is widely perceived to be helpful in revealing communities and thereby behavioral patterns. Could there be a correlation between preferences shown by members to certain geographical sites and some of the other common features of a social network’s members? What role do demographic features play in determining location-based preferences of members?

The author uses two datasets to derive the features. The first one pertains to users and their check-in records in London, and the second one pertains to users and their check-in records in Berlin. A friendship model is proposed where a connected graph is visualized with nodes forming the users in the social network, the edges revealing a common topic, and the weight of the edges revealing the extent of participation of the connected users.

A multi-label support vector machine (SVM) classifier is then trained with a set of features that include user interests and demographic features derived from the friendship model. Ninety percent of the dataset is used for training the SVM classifier, while the model is tested on the remaining ten percent. Precision and recall figures are analyzed for both the datasets to highlight the utility of prediction, even for imbalanced datasets. The influence on demographic features for communities in different locations is also analyzed by training the classifiers on datasets separately, with and without those features.

This work is likely to generate interest in agencies associated with deploying social networks, as it can suggest targeted advertising to location-based service providers like restaurants, transportation providers, educational institutions, and so on, by recommending prospective users as targets.