A community division algorithm integrating influence and label propagation
Since traditional label propagation algorithms adopt random strategies during node update sequence initialization and label update,their community division results usually face instability and low accuracy.In regard of this,we design a community division algorithm integrating influence and label propagation,namely ILPCD.First,the network is hierarchically divided by the k-shell algorithm.The global influence of each node in the network is calculated by its k-shell value and its first-order neighboring nodes'k-shell values.Then,the node update sequence is initialized in descending order of the global influences,thus the randomness of the update sequence can be eliminated.Second,the calculation method for Jaccard similarity coefficient is improved by introducing a smoothing coefficient,thus the correlation between nodes can be more objectively measured and the nodes'local influences can be obtained.Finally,labels are updated according to the local influences,and the randomness in the process of label propagation can also be eliminated.Experimental results on four real network datasets show that the ILPCD algorithm can effectively divide communities,and has higher stability and accuracy.
community divisioninfluencek-shell algorithmsimilarity coefficient
万方数据
维普
