Topluluk dinamikleri ve kişisel öneriler: Tek parçalı ve çift parçalı ağlarda bağlantı tahmininin iyileştirilmesi

Complex networks serve as suitable models to represent systems involving interactions between entities. The dynamics of these interactions often lead to the formation of communities. Network modeling can be leveraged to tackle a number of tasks, including link prediction, which allows predicting future interactions between entities. Communities can be defined as subsets of the network where the links between nodes are more dense. Therefore, we expect nodes within the same community to have a higher tendency to link with each other. Yet, the dynamics of the relationship between link prediction and community structure remain understudied in the literature. Additionally, there is a lack of comparative studies on link prediction methods for bipartite networks. Our aim is to delve into the link prediction phenomenon, first by revealing its relation to community structure in unipartite networks and, second, by establishing a systematic methodology for improving prediction accuracy in bipartite networks. We present a series of experiments for the former and four link prediction strategies adapted to work in bipartite networks for the latter. Our findings are two-fold. First, the more pronounced community structures in a network, the better the link prediction success, regardless of the identification of the communities. Moreover, performing the link prediction task on a per-community basis improves link prediction success. Second, adaptation of state-of-the-art personalized collaborative filtering methods can perform link prediction successfully in bipartite networks. Lastly, the SPM link prediction method, originally developed for unipartite networks, also demonstrates relatively high accuracy at predicting links in bipartite networks.