Study Of Density-based Semi-supervised Clustering Algorithm On Complex Network

Complex network is a network structure that contains a large number of nodes and edges between the nodes and it is widespread in nature and society. Research shows that community structure is one of the most important properties in complex network. Complex network clustering is aimed to find the community structure in the network. It can help us understand the intrinsic property of the network and deal with the real system.Firstly, we made a research on density based clustering algorithms. For the problem that most of existing density based clustering algorithms cannot make use of the prior information effectively, we provided a density-based semi-supervised clustering algorithm. The prior information was organized as constraint pairs in this algorithm and we expanded the constraint pairs with their transitivity and symmetry. We could find the connected and largest community structure under the guidance of the expanded prior information. The algorithm can improve the accuracy of the result.Then, we made a research on semi-supervised clustering algorithms. For the problem that prior information is hard to get by itself, we provided density-based active learning algorithm to generate valuable prior information. The algorithm chose the active nodes by comparing the similarity of nodes. The chosen least active nodes can cover all the community and control the border. We brought the active prior information into the original density-based semi-supervised clustering algorithm. It can improve the clustering performance of original density-based semi-supervised clustering algorithm.At last, we tested our algorithms with experiments. We compared our density-based semi-supervised clustering algorithm with other two semi-supervised clustering algorithms on both real network and LFR network. The result shows that the density-based semi-supervised clustering algorithm has a better performance on both accuracy and time efficiency than other two algorithms. When we brought active learning into semi-supervised algorithm, it improved the accuracy of the original density-based semi-supervised clustering algorithm.