Improved LFM algorithm in weighted network based on rand walk

Because of randomly selection of seed nodes, the result of traditional LFM algorithm is full of instability. What's more, with underused weight information in network, the accuracy of LFM decreases apparently in network with fuzzing community structure. In order to solve the problems, LFMs algorithm is presented in this paper. First, the random walk method was used to select seed nodes to avoid the instability of LFM. Then, with cosine similarity to calculate vertex similarity, weight information in network was fully used, and the precision of community division was also raised. To validate the algorithm, LFMs was compared with traditional LFM in LFR benchmark and real network. Results showed that, both in LFR network and real network, LFMs gets higher precision than LFM.