Link prediction in multilayer social networks using reliable local random walk and boosting ensemble classifier

This paper presents an enhanced approach for predicting links in social networks by utilizing a Boosting Ensemble Classifier, and Reliable Local Random Walk (BEC-RLRW). Existing methods often fall short in capturing the complex dynamics and inter-layer relationships inherent in multilayer social networks. By integrating reliable LRW with boosting ensemble classifier, our approach aims to address these shortcomings by providing a more reliable similarity metric and a robust classification model. BEC-RLRW creates a novel transition matrix based on a similarity metric based on reliable local random walk. Metrics that convert unweighted to weighted similarity can be effectively created by establishing trustworthy and reliable paths between nodes. Additionally, a popular method for estimating linkages in weighted multilayer networks is the local random walk. The purpose of BEC-RLRW is to develop a reliable local random walk as a multiplex similarity metric in multilayer social networks. In the next step, the features of nodes are extracted based on node2vec embedding and the results are used for edges embedding. When paired with the corresponding positive or negative labels, the resulting edges embedding creates a well-labeled dataset that can be used for link prediction. Eventually, a set of potential edges are identified by applying the well-labeled dataset to a boosting ensemble classifier. To ensure the optimal performance of the proposed algorithm for link prediction in multilayer social networks, we conducted extensive experimental tests on several real-world networks. The obtained results show the efficiency and performance guarantee of our method compared to the existing methods.