Community Detection Algorithm Research In Bipartite Networks

Bipartite network is a kind of important forms of complex network. Thecommunity of the network was a lot of nodes which owns much inside-line morethan outside-line. Detecting the communities in the network is of great significancefor learning and analyzing the characteristics of the network. The method ofcommunity detection in bipartite networks started with projection algorithm, itprojects the number of the network’s node type into one single and then makesfull use of the method which is so mature in single networks to detect thecommunities. But then it pointed out that this method is not precise, so the methodsthat directly making the community detection on the bipartite networks have beenproposed. Barber’s BRIM algorithm is one of them which was based on the matrixof the single network module and expanded to the bipartite networks. The BRIMmade an effective division, but it need for additional input parameters in theinitialization phase, which was a greater limitation when making application.We present a new agglomerative algorithm--MAB which is based on themodularity of bipartite networks and parameter-free. It treats each node as aseparate community and agglomerates the communities by maximizing themodularity of the networks, and then applies the algorithm to real-world networkdata and compare with the other methods that process on this, showing that thealgorithm successfully identifies the modular structure of bipartite networks. Themethod of MAB calculated the increment of the networks’ modularity when it triedto merger any two of the communities and finally merged the maximal couple. Butevery step has to traverse the whole network which takes more time. We propose amethod that based on the heap of modularity, it is not only of no need forparameters when successfully detecting community but also reducing the algorithmcomplexity.