Research On Community Discovery In Association Network

Association network is the abstraction of entities and their relationships.It's widely used in the study of many real world systems.Community,also mentioned as model, group etc.is a common phenomenon in association network,which exists as a reduction of the whole network.Discovering communities form dynamic network and detecting their evolution process is of great importance in analyzing topology features and dynamic characteristics.Recent years witness an increasing interest in the study of community discovery.So far,community techniques have been applied to network optimization,resource discovery and resource recommendation.However,existing community discovery methods have the following limitations.Firstly,the extracted communities are of the same level,either disjoint or partly overlapping,while communities tend to be hierarchical,i.e.a large community may further fall into several sub-communities.Due to the lack of community information,it is hard to get a full outline of the whole network and further effectively detect the changes of communities.Moreover,the existing classification of evolution categories can't satisfy the complexity of possible variation.To overcome above problems,this thesis proposes an approach to community discovery and applies it to word association network from DBLP bibliography titles.The research work includes the following aspects:Firstly,briefly introduces and analyzes the state of arts;then presents a framework for community discovery in association network:gives the formal description of association network with temporal factor,association network segment,community hierarchy structure and community evolution graph. Sccondly,details how to discover hierarchical communities and identify community evolution process in association network,including basic idea of community hierarchy on cohesion and association level ontology,algorithms for community hierarchy construction,classification of evolution categories,metrics of evolutionary degree,and algorithm to identify evolution process.Thirdly,constructs a word association network from DBLP bibliography dataset;uses the proposed method to discover its community structure and analyzes the re.sult.The outcome suggests that this approach provides a feasible and effective method for community discovery.