Evaluation of Graph-based Algorithms for Guessing User Recommendations of the Social Network Instagram

In this paper, we present an analysis of the user recommendation algorithm of the popular social network Instagram. First, we introduce a framework for the analysis and evaluation of user recommendations in any social network. Then, we build a neighborhood graph for selected active users on Instagram, containing more than 609,000 users and 747,000 relationships. For these users, we also create a dataset of not publicly available personalized recommendation lists. Finally, we apply several graph-based algorithms to this graph to identify Instagram's user recommendation system. We show that an algorithm based on Twitter's Who-to-Follow algorithm is most similar to Instagram's unknown algorithm when recommending users connected to the followees of the target users. Moreover, this algorithm is very cost-effective due to requiring only a relatively small, bipartite graph, compared to the other candidates. Furthermore, we give an analysis of the composition of the user suggestions based on the different reasons for the suggestions.