Performance of Common Classifiers on node2vec Network Representations

In this paper we evaluate the performance of different multi-class classifiers on network graphs. Since the node embedding techniques have been widely used to represent and analyze networks structures, we decide to transform network data (nodes and links) into attributes which are descriptive and contain correct information of its structure. For this purpose, we use a state-of-the-art algorithmic framework node2vec, which has been shown to outperform other popular methods when applied to multilabel classification as it manages to efficiently learn a mapping of nodes to a low-dimensional space of features. Applying this framework, we generate a set of representations for nodes of multiple network data sets. Using the generated representations, we evaluate the performance of common classifiers. We perform crossvalidation and parameter tuning to get the best possible model of each classifier type. To compare their performance, we computed Precision, Recall and F1-score for each model on each data set. Following that, the obtained results are analyzed and compared.