Research On Algorithm Of Graph Mining Based On Attribute Fusion

Since the rapid development of the Internet,social networks and people's daily life become affluent.The structure of the data presents more and more complex,and graph data has arises now.The graph data includes attributes and the topological structure between the different data object.In order to dig out knowledge and information from the massive graph data,the researchers begin to expensive study the graph clustering technology with node attributes.As a new research direction in the field of data mining,the graph clustering technology has important practical research significance.Most current graph clustering algorithms are only based on node topology or the attributes of the node.The rise of social networks makes these graph clustering algorithms based on a single feature unable to meet the growing needs of person.Graph clustering algorithms only based on a single feature cannot have a better clustering effect,and it is not accurate to model the actual problem or reflect the actual distribution of the data.Therefore,digging attribute graph has important research value and practical significance.Now the graph clustering algorithm only considers the topological structure of graph data or the attributes of graph data.This thesis proposes a multi-layer attribute fusion model to solve the problem which only considers the topological structure or the attributes.Through the attribute fusion model strategy,the graph data will be modeled different levels as the attributes level and the topological structure level,and the different weight level coefficients are set according to the importance of the object attributes.Finally,according to the model fusion strategy,the attributes of the graph data and the relationship between the different data object integration into a low-level network.Through the analysis,the model fusion method can more reflect the data actual distribution.In this thesis,an adaptive clustering algorithm based on multi-layer attribute fusion for the Adaptive Weight Distribution(MAFAWD)algorithm is proposed to solve the problem that the existing graph clustering algorithm is not ideal for graph data clustering.The algorithm first establishes a graph model of the data,and then divides the different attributes layers and structure layer.By setting different attribute layer and structural layer weight coefficients,the data modeling is more reflected the actual distribution.Then the thesis uses Affinity Propagation Clustering algorithm to cluster the graph data.In order to achieve the ideal clustering effect,the coefficients of the attribute layer is changed adaptively according to the node voting mechanism,so that the vertices within the same cluster are closely related and have the same attribute,in the different cluster vertex attributes different and the edges are sparse.Finally,this thesis verifies the MAFAWD algorithm on the DBLP real data set.Through the experimental simulation design and the comparison and evaluation of the clustering results,it is shown that this thesis proposes the MAFAWD graph clustering algorithm is superior to the current graph clustering algorithm.