Research On Multiple Relation Social Network Community Detection Base On Subspace Clustering

With the development of information technology in the past few years, people have become more powerful in the ability to achieve and store data. Vast amounts of data often contains a lot of interesting information that will usually reveal some potential realworld laws. Data mining is to meet people’s demand of mining potential information from massive data generated. Clustering algorithm as a branch of data mining in recommendation, banking credit risk early warning, Quantitative trading has a very widely applications. While data becomes more abundant and dimension becomes more complex,conventional clustering algorithms are difficult to get good results. To solve the problem of high-dimensional data clustering, subspace clustering is proposed and a widely applications. Through the selection and conversion of data dimension, subspace clustering can detect hidden in the high-dimensional data subspace community structure. This paper studied multiple relation social network, community detection base on subspace clustering.The main work of this paper includes the following aspects:1) It introduces the basic concepts of clustering algorithms and general process. And It analyze the advantages and disadvantages of each method which are usually used in dimension reduction, similarity measure, and evaluation of clustering results. Then, based on the principle of clustering algorithms, it introduces several representative clustering methods, such as hierarchical clustering algorithm, partition-based clustering algorithm,spectral clustering algorithm, subspace clustering algorithm, and the advantages and disadvantages of each algorithm.2) The second mission is to study the community detection algorithm which based on sparse subspace clustering. This method obtains similarity matrix by calculating the sparse representation of each point base on compressed sensing theory. Then, it remove the random factor and common factor from similarity matrix base on random matrix theory. Finally, divide the data set into communities by using modularity-based community detection algorithm. To verify the effect of the algorithm, this article has tow experiments on social networks and financial networks. Comparing with spectral clustering and Newman fast clustering algorithm, sparse subspace clustering community detection algorithms which not only has advantage of some but its partition is more aligned with the ground truth of data set. It shows that sparse subspace clustering community detection is superiority and more effectiveness on community detection in different data set.3) Introduce the design and verification of multi-relation social networking community detection system. System consists of data collection and pretreatment module, data analysis module, data structure display module and so on. And It contains the multisource heterogeneous data extraction, community detection, and results analysis exhibition. This paper describes the system in the overall structure of the design, the main partition of design of each module and module verification. The main features of this system are : In order to be better able to meet the multi-source heterogeneous societies inspection requirements. The system use of design patterns that it shows greater flexibility and scalability to make the system to meet the open- closed principle, the access multiple data sources, and more integration algorithm.