Hybrid Link Prediction Model with Explainable AI for Online Social Networks

Online social network systems are inherently complex due to the various types of connections that can form between any two nodes within a single network. These intricate structures enable users to engage in diverse interactions, resulting in constantly evolving behaviors. This results in cross dimensional evolution of complex networks with links forming between social entities with variable tie strength. This makes prediction of links in online networks quite significant for several higher tasks, like knowledge extraction, network learning, signed behavior analysis, etc. Though there exist several link prediction scores, however, the existing literature lacks hybrid link prediction score to compute the possible formation of strong or weak ties in a social network. In this paper, a hybrid prediction mechanism, namely GML-hybrid is proposed to estimate the possible links using network theory and machine learning. The proposed hybrid link prediction algorithm combines already existing distance metrics and extracts different thresholds across three different levels which comprise of first, second and third quantile used for dataset labelling. This proposed algorithm is applied on Ego Facebook dataset considering 15 classifiers for model building Further, the results obtained from 5 highest performing models are further analyzed using explainable AI using SHAP.