Efficient Distributed Clustering Algorithms on Star-Schema Heterogeneous Graphs

Many datasets including social media data and bibliographic can be modeled as graphs. Clustering such graphs is able to provide useful insights into the structure of data. To improve quality clustering, node attributes taken account, resulting in attributed Existing graph clustering methods generally consider attribute similarity structural separately. In this paper, we represent star-schema heterogeneous graphs, where are different types nodes. This enables use personalized pagerank (PPR) a unified distance measure that captures both similarities. We employ DBSCAN for update edge weights iteratively balance importance attributes. The rapidly growing volume nowadays challenges traditional algorithms, thus, distributed method required. Hence, adopt popular computing system Blogel, based on which, develop four exact approximate approaches enable efficient PPR score computation when updated. effectiveness propose simple yet effective weight strategy entropy. addition, present game theory trading efficiency result quality. Extensive experiments real-life offer our proposals.