Work

大規模グラフに対するクラスタリングの高速化に関する研究

Year: 2015
Type: dissertation
Abstract: Recent advances in social and information science have provided linked data that model our society and the natural world around us.Since graphs are fundamental data structures that can naturally represent data entities as well as the relationships among entities, their importance is increasing in representing complicated structures and schema-less data such as data generated by Twitter, Facebook and various complex networks.For these complex networks, graph cluster analysis is one of the most important techniques in various research areas such as data mining, social science and marketing.The problem of the graph cluster analysis is to find clusters of nodes which are densely connected within the cluster and sparsely connected among clusters, and this problem has been studied for some decades in many fields.Since the size of graphs is increasing significantly in the big data era, efficient clustering algorithms that can handle large-scale graphs are highly demanded.In order to achieve efficient clustering of large-scale graphs, this dissertation investigates three novel algorithms taking statistical properties of real-world graphs into consideration: (1) we propose an incremental agglomerative graph clustering algorithm for modularity-based clustering, (2) we investigate a data parallel clustering method for modularity-based graph clustering by using SIMD instructions, and (3) we introduce a novel clustering algorithm for structural clustering by handling two-hop away nodes in the real-world graphs for efficient clustering.As a result, in the experiments over real-world and synthetic datasets, this dissertation proves that the algorithms can achieve not only efficient clustering but also highly accurate clustering results for large-scale graphs.By providing efficient and sophisticated approaches that suit for large-scale graphs, the algorithms will help in increasing the effectiveness of clustering in a wide range of applications.
Language: English
Open Access status: green
OpenAlex ID: W7145897999
Cites: 118
Cited by:
Related to: 10