Summary: | 碩士 === 國立交通大學 === 電信工程研究所 === 101 === With the rise of social networking service, operators are very interested in the data analysis of social network. However, due to the large data size, traditional algorithms running on single computer cannot deal with it. Thus, how to design an algorithm in MapReduce is an important issue. The thesis focuses on designing algorithms for finding cohesive subgraph (k − truss) in MapReduce. We propose two algorithms applied to two situations. The first one is designed for static social networks. In other words, it can analyze a social network at a snapshot. The second one is designed for dynamic social networks. Given two close snapshots of a social network, we use the analytical results of the first one to speed up the analysis of the second one. We test the proposed algorithms by both synthetic data and real world data including email communication network, online social network and so on. The first algorithm is faster than the original one when applied to real world data. In case the proposed algorithm does not improve the performance, our
algorithm can automatically switch to the original algorithm. We verify the second algorithm using synthetic data. It is faster than the original one if the change of networks is below a threshold.
|