Extracting Communities in Networks Based on Functional Properties of Nodes

Discriminating User Behavior though PC Operation Logs by PageRank Convergence Patterns

“…The functionality clustering method by Fushimi et al [5] regards the convergence process of the PageRank score of each node as a time series vector. Next, these vectors are clustered by the greedy k-median method with cosine similarity between vectors.…”

“…Thus, we need to take into account of delay and scale of patterns. As a measure of similarity between convergence patterns, the functionality clustering [5] employs a cosine similarity. The cosine similarity is relatively robust against the scale of patterns, while susceptible to the delay of data.…”

“…2) An implicit approach (by Fushimi et al [5]). In this method, the similarity between convergence patterns of nodes in the PageRank algorithm is regarded as the similarity between the nodes.…”

“…In this method, the similarity between convergence patterns of nodes in the PageRank algorithm is regarded as the similarity between the nodes. The convergence pattern of a node implies the structure in the vicinity of the node, which is called functionality in [5]. Hereafter, we refer to this method as functionality clustering.…”

“…Hereafter, we refer to this method as functionality clustering. For this paper, we employed the implicit approach by Fushimi [5] to measure the similarity between two nodes according to the roles of the nodes in the network. Fushimi's method employs the cosine similarity to measure the distance of the convergence patterns of nodes.…”