Estimating Node Characteristics from Topological Structure of Social Networks

Kouhei SUGIYAMA  Hiroyuki OHSAKI  Makoto IMASE  

IEICE TRANSACTIONS on Communications   Vol.E92-B   No.10   pp.3094-3101
Publication Date: 2009/10/01
Online ISSN: 1745-1345
DOI: 10.1587/transcom.E92.B.3094
Print ISSN: 0916-8516
Type of Manuscript: PAPER
Category: Fundamental Theories for Communications
complex network,  social network,  link mining,  SSI (Structural Superiority Index),  prospect theory,  

Full Text: PDF(435.4KB)
>>Buy this Article

In this paper, for systematically evaluating estimation methods of node characteristics, we first propose a social network generation model called LRE (Linkage with Relative Evaluation). LRE is a network generation model, which aims to reproduce the characteristics of a social network. LRE utilizes the fact that people generally build relationships with others based on relative evaluation, rather than absolute evaluation. We then extensively evaluate the accuracy of the estimation method called SSI (Structural Superiority Index). We reveal that SSI is effective for finding good nodes (e.g., top 10% nodes), but cannot be used for finding excellent nodes (e.g., top 1% nodes). For alleviating the problems of SSI, we propose a novel scheme for enhancing existing estimation methods called RENC (Recursive Estimation of Node Characteristic). RENC reduces the effect of noise by recursively estimating node characteristics. By investigating the estimation accuracy with RENC, we show that RENC is quite effective for improving the estimation accuracy in practical situations.