Research On The Seed Set Expansion Algorithm Of Complex Networks

Complex network is an effective tool for modeling and analysis of associated data.In many complex network application scenarios,such as online social networks or recommendation systems,there is a lot of interest in identifying a group of similar users/nodes/items.However,it is obviously unrealistic to manually identify large networks.An effective method is to label some nodes as seed nodes,and to expand it.That is seed set expansion.However,how to effectively expansion using limited information that seed members provide become a very challenging topic.In the existing research work,this project aims to provide a completely new perspective for seed set expansion,and compares the existing algorithm with the network data based on real communities to verify the effectiveness of the algorithm.First of all,Using the theory of propagation dynamics of complex networks to analyze the evolution process of communities in network,we transform seed set expansion problem into the label propagation problem of propagation dynamics,and propose a Soft Label Propagation algorithm and Augmented Label Propagation considering similarity.Secondly,borrowing the idea that the feature space distance represents similarity,based on the theory of network representation learning and machine learning,the high-dimensional network adjacency matrix is mapped to the low-dimensional feature space,and the seed set expansion problem is turned into finding nearest neighbors in the feature space,and firstly propose a KNN-like algorithm based on Deep Walk.Then,Based on the model and network interactive learning mode to expand seed set,associating similar reinforcement learning.The seed set expansion problem is compared with a common scenario maze path finding problem of reinforcement learning,and a Q-Learning-based algorithm is designed for seed set expansion.We compared the existing algorithms with the network data based on real communities and verified the effectiveness and effectiveness of our algorithm.