Link prediction analysis based on Node2Vec embedding technique

Abstract

The paper focuses on analysing link prediction using the Node2Vec embedding technique, which is based on the Random Walk algorithm. In addition to this, several machine learning models have been employed to assess the effectiveness of the embedding technique. Node2Vec employs various embedding operators, including Hadamard, Concatenation, Average, Weighted L1, and Weighted L2. The comparative analysis of this embedding technique is done on real world network data sets using various machine learning models with state of the art link prediction algorithms. Performance assessment of Node2Vec's embedding technique is based on the AUC metric. According to the simulation results, it has been determined that the concatenation operator with the bagging classifier yields mean AUC value of 0.939, outperforming the other operators, which produce AUC values below 0.91. Furthermore, the study has also revealed that the embedding technique provides superior results when applied to networks with a low ratio of nodes to edges.