Within-Network Classification Using Local Structure Similarity

Abstract

Abstract. Within-network classification, where the goal is to classify the nodes of a partly labeled network, is a semi-supervised learning problem that has applications in several important domains like image processing, the classification of documents, and the detection of malicious activities. While most methods for this problem infer the missing labels collectively based on the hypothesis that linked or nearby nodes are likely to have the same labels, there are many types of networks for which this assumption fails, e.g., molecular graphs, trading networks, etc. In this paper, we present a collective classification method, based on relaxation labeling, that classifies entities of a network using their local structure. This method uses a marginalized similarity kernel that compares the local structure of two nodes with parallel random walks in the network. Through experimentation on different datasets, we show our method to be more accurate than several state-of-the-art approaches for this problem.

Within-Network Classification Using Radius-Constrained Neighborhood Patterns

“…A state-of-the-art structure-aware WNC method called RL-RW-Deg [8] 2 was employed as the baseline. We aim to empirically answer the following questions: 1) Does neighborhood features consistently outperform the state-of-the-art?…”

“…However, none of those classifiers exploits the rich structural information of the network. [8] proposes RL-RW-Deg, which characterizes the structure of a vertex's neighborhood by its degree and the distribution of different label sequences obtained after starting a random walk from the vertex. RL-RW-Deg is reported to outperform various non-structural methods [23,28,22] in the chemical structure completion task.…”

“…In Figure 3, we use the chemical structure completion task [8] to compete our approach with the baseline on testing networks of different label ratio…”

“…Instead, we relate the classification of a vertex to the structure of its neighborhood, which is not limited to the information of all neighbors' labels. The "homophily-or-structure" issue has caught some attention in recent WNC studies [12,8,25]. But to the best of our knowledge, we are the first to address this issue under the context of frequent-pattern-based classification.…”

MapReduce approach to relational influence propagation in complex networks

“…Among others, several statistical relational learning (SRL) techniques were introduced, including probabilistic relational models, relational Markov networks and probabilistic entity-relationship models [7,8]. Two distinct types of classification in networks may be distinguished: based on collection of local conditional classifiers and based on the classification stated as one global objective function.…”