The Influence of Network Structural Preference on Node Classification and Link Prediction

Recent advances in complex network analysis opened a wide range of possibilities for applications in diverse fields. The power of the network analysis depends on the node features. The topology-based node features are realizations of local and global spatial relations and node connectivity structure. Hence, collecting correct information on the node characteristics and the connectivity structure of the neighboring nodes plays the most prominent role in node classification and link prediction in complex network analysis. The present work introduces a new feature abstraction method, namely the Transition Probabilities Matrix (TPM), based on embedding anonymous random walks on feature vectors. The node feature vectors consist of transition probabilities obtained from sets of walks in a predefined radius. The transition probabilities are directly related to the local connectivity structure, hence correctly embedded onto feature vectors. The success of the proposed embedding method is tested on node identification/classification and link prediction on three commonly used real-world networks. In real-world networks, nodes with similar connectivity structures are common; Thus, obtaining information from similar networks for predictions on the new networks is the distinguishing characteristic that makes the proposed algorithm superior to the state-of-the-art algorithms in terms of cross-networks generalization tasks.