Social Network Analysis Based On Powergraph Parallel Computing Framework

Graph is a fundamental data structure that captures relationships between different data entities.In practice,graphs are widely used for modeling complicated data in different application domains such as social networks,protein networks,transportation networks,knowledge bases and many more.Nowadays,the relationships between people,communities,organizations,countries and other actors are becoming more and more close,and the valuable information in these relationships also grows rapidly,which make the research of social network analysis increasingly hot.In principle,social network analytics is an important big data discovery technique.Currently,large-scale social networks with millions and billions of nodes and edges have become very common.In order to process and analyze the large-scale networks,there have appeared a number of distributed graph parallel computing platforms that meet the characteristics of large-scale network computing.However,many classical algorithms of social network analysis are centralized algorithm based on one computer design,can not satisfy the needs of large-scale social network analysis.Therefore,this paper focuses on the three aspects of social network spreading,matrix decomposition and network rewiring.The parallel graph data analysis algorithm is designed and implemented under the PowerGraph parallel computing framework.This paper mainly completed the following aspects of the work:1)Parallel spreading algorithm based on PowerGraphThe spread of the virus,the diffusion of information,etc.,can be seen as obeying a certain kind of network spreading behavior.Through the spreading model,we can imitate these spreading behaviors,which help people understand the spreading mechanism.For different viruses or information,the applicable spreading model is not the same.The classic spreading models are SIS,SIR,SIRS and SEIR.The Parallel Spreading Algorithm for SEIR Model(PSA-SEIR)based on PowerGraph is proposed.The experimental results show that the simulation results are consistent with the spreading trend of SEIR spreading model,and the scalability of the algorithm is analyzed.2)Parallel matrix decomposition(SVD ++)algorithm based on PowerGraphIn the social network analysis,matrix decomposition is a common method.A network can be abstracted as a matrix,the social network analysis algorithm can be solved in matrix computing.Therefore,an effective decomposition of large-scale sparse matrix can solve many practical problems(eg,movie recommendation).Based on this,the original parallel SVD ++ algorithm is improved,and the Learning Rate Adjustment Parallel SVD++ Algorithm(LRA-PSVD++)based on PowerGraph is realized.It is verified that the accuracy of LRA-PSVD++ algorithm is increased by experiment.The scalability of the algorithm is analyzed.3)Parallel rewiring algorithm based on PowerGraphThe macro topology of the network has a lot to do with its many basic characteristics.Assortativity is an important feature of network macroscopic topology,and the change of Assortativity implies the change of network topology.Through network rewiring,simulating networks with different Assortativity coefficients can help to analyze other basic characteristics of different networks(eg,spreading characteristics,robustness).Based on this,the Parallel Random Rewiring Algorithm(PRRWA)based on PowerGraph is proposed under the condition of keeping the degree sequence as the goal of increasing Assortativity.The feasibility and scalability of the algorithm are analyzed by experiment.