Application and analysis of link prediction in academic networks

Nowadays, along with improvements in the Internet and social networks, social network analysis and link prediction are still popular. To solve the link prediction problem, link prediction methods used in web mining are used together with graph structures. In addition, graph structures come into prominence for modeling real-world problems. Social network analysis methods are applied to these models, and inferences are made about properties of the actors corresponding to the node in graph and the situation of inter-actor relations. In this thesis, the parameters and methods that can be used primarily for social network analysis and link prediction were explained with examples and network models that can be analyzed were mathematically examined. Then, the software on link prediction was performed in the academic networks. Author-article network in software, modeled as a binary network, the authors were considered as the main nodes. There are links between authors who co-publish in the software, and these links are represented by the edges. In the software, the keywords of the articles were taken as reference. There are also links between the authors and the keywords of the articles that they write. By calculating the similarities between the authors and the keywords that they are related to, attempts were made to predict possible new connections between the authors. By considering the topics that the authors want to study, it is observed that the authors who can work together in an area where an author wants to work are tried to be determined and it is observed that it can be achieved with the obtained results. It has been worked on a known network, for the accuracy of the software. It has been gone to the time before some connections, and the probabilities of the connections occurring today were calculated and very good results were obtained.