Identifying Missing Node Information in Social Networks

Abstract

In recent years, social networks have surged in popularity as one of the main applications of the Internet. This has generated great interest in researching these networks by various fields in the scientific community. One key aspect of social network research is identifying important missing information which is not explicitly represented in the network, or is not visible to all. To date, this line of research typically focused on what connections were missing between nodes,or what is termed the "Missing Link Problem." This paper introduces a new Missing Nodes Identification problem where missing members in the social network structure must be identified. Towards solving this problem, we present an approach based on clustering algorithms combined with measures from missing link research. We show that this approach has beneficial results in the missing nodes identification process and we measure its performance in several different scenarios.

A Hybrid Approach of Classifier and Clustering for Solving the Missing Node Problem

“…We then formally define the problem which we address, describe how HACC is implemented in this domain, and our experimental setup. Overview and Background We focus on a specific variation of the missing nodes problem where the missing nodes requiring identification are "friends" of known nodes, as previously done with the MISC and SAMI based algorithms (Eyal, Rosenfeld, and Kraus 2011;Sina, Rosenfeld, and Kraus 2013). An unidentified friend is associated with a "placeholder" node to indicate the existence of the missing friend.…”

“…The missing node identification problem has recently generated research interest. Previously, Eyal et al (Eyal, Rosenfeld, and Kraus 2011;Eyal et al 2014) presented the MISC algorithm in order to solve this problem. MISC com-Figure 1: A full network, the known network and the visible network obtained (right) by adding the placeholders for the missing nodes 1 and 5, which need to be united into clusters.…”

“…We considered as input a variety of known similarity measures as fields within the vectors. These include features based on network structure that were previously studied (Eyal, Rosenfeld, and Kraus 2011). Specifically, we used the two structure similarity measures that previously yielded the best results: the Relative Common Neighbors (RCN) measure (Liben-Nowell and Kleinberg 2007) and the Adamic/Adar (AA) measure (Adamic and Adar 2003), and a third structure similarity measure which is the combination of the two measures (RCN+AA), formally defined as…”

“…Note that although MI-Alg is a generalization of the MISC algorithm (Eyal, Rosenfeld, and Kraus 2011), it also introduces a major change in addition to HACC's novel calculation of the affinity matrix. In contrast to MISC, MI-Alg no longer processes the affinity measures of missing nodes through spectral clustering but instead exclusively uses the K-means clustering algorithm on the placeholders' affinity matrix.…”

“…HACC creates one combined affinity measure between each object pair, which is the classifier's confidence score, that measures the probability that two objects needing to be clustered will be in the same cluster. We first demonstrate how HACC can be implemented in order to solve the missing node identification problem (Eyal, Rosenfeld, and Kraus 2011), which locates and identifies missing nodes within the network. For this domain, we conducted a thorough study which includes an intensive evaluation and analysis of 5 different types of classifiers and 12 different sets of domain features.…”