Research And Implementation On Graph OLAM Large-scale Multidimensional Network Analysis And Mining System Based On Distributed Platform

Over years,with the rapid development of Internet information technol-ogy,massive amount of data has been accumulated from real-world applica-tions.Network has been widely adopted for modeling complex datasets in dif-ferent application scenarios.Such networks always contain a lot of dimensions,which makes it challenging to extract useful information from them.How to distill valuable information from multidimensional network to support decision making has attracted considerable attention.Inspired by traditional OLAP technology,Graph OLAP(Graph On-line Analytical Processing)technology are proposed to enable users to view multi-dimensional network data from different angles and different granularity lev-els.However,the existing works on Graph OLAP always center on studying the graph cube model,and most of the operations are inherited from traditional OLAP technology.They fail to exploit the topology structure of multidimen-sional network to mine in-depth information.On the other hand,the traditional graph mining technology based on graph theory focuses on mining the relation-ship between nodes and edges and describes the characteristics of the network by extracting the hidden information from topology structure.Although some works adopted it for the preprocessing of Graph OLAP,they do not discuss how to mix them together to extract more in-depth information.In this paper,the concept of Graph OLAM(Graph On-line Analytical Mining)which combines traditional Graph OLAP technology and graph mining technology is proposed to support more complex and in-depth analysis of multidimensional network data.The main research can be summarized as follows:(1)The concepts of Path Aggregate Network and Dimension Aggregate Network are formally defined,and a new graph data cube model is proposed to guide the aggregations of multi-dimensional networks.The proposed graph cube model has better performance than the existing models.(2)The concept of Graph OLAM operations is formally introduced which can be constructed as Graph OLAM workflow.The proposed Graph OLAM operations are categorized into two major subcases.(3)According to the concept of Graph OLAM operations,a large-scale multi-dimensional network analysis and mining framework is proposed.The optimization strategies are discussed around the materialization strategies of graph data cube and the optimization strategies of the analysis framework.(4)The framework is implemented based on the distributed storage plat-form and parallel computing framework.Extensive experimental evaluations on large-scale real and synthetic data set are conducted.The experimental re-sults demonstrate the superiority of our proposed Graph OLAM framework,which proves that the framework can analyze multi-dimensional network more effectively and efficiently.