Research On Robust Semi-supervised Community Detection Method

With the further development of complex network research,community structure has become a research hotspot.People realized that identification of community structure can help them understand and use complex networks better.In recent years,there have been many high quality research results in the identification of community structure,but few researches on noise in communities cannot guarantee the robustness of the algorithm.Therefore,how to identify and process the noise existing in the network has become an urgent problem to be solved under the premise of guaranteeing the quality of community discovery algorithm.Aiming at solving the above problems and challenges,based on the analysis and research of classical algorithm existing in semi-supervised community detection,we respectively analyzed the effect of noise on semi-supervised community detection methods under the condition of two kinds of prior knowledge which are individual labels and pairwise constraints,and finally studied the method to identify and remove noise.The innovative results are as follows:(1)For the shortcomings in the present study that semi-supervised community detection methods did not consider the effect of noise generated by the network,this Dissertation proposes a harmonic function of semi-supervised community detection method in order to identify the noise existing in the network,which can help semi-supervised community detection method use prior knowledge more effectively,and then improve its accuracy significantly.In this strategy,a harmonic function is embedded in the objective function to deal with the differences between nodes and its neighbors.By sorting those differences,nodes in different communities in the network can be obtained as pseudo noise nodes.(2)In view of the two kinds of existence prior knowledge individual labels and pairwise constraints in semi-supervised community detection methods,we considered the characteristics of each prior knowledge respectively and removed the noise identified in previous work in prior knowledge.Combining the classic algorithm SLPA,SDPT and SSNMF,we conducted experiments on artificial and real datasets.The experimental results proved that removing the noise from prior knowledge can improve the robustness of semi-supervised community detection algorithm.In this Dissertation,we studied the noise reduction strategy and robustness of semi-supervised community detection method from two aspects of class labels and pairwise constraints respectively and put forward the corresponding solutions,and we also showed that this study is effective to explore and expand the methods in the field of semi-supervised community detection through theoretical analysis and simulation results.