Research On Community Member Identification Technology Based On Ensemble Learning

In the reality,many systems can be modeled as corresponding complex networks for analysis,such as microblogging networks,protein networks,and literature networks.These networks are usually composed of modules(also known as communities),with high connectivity inside the modules and lower modules.Community detection is an important issue in complex network analysis.It has importantly theoretical and practical value for discovering the hidden relationship between nodes in the network and mining network information.The traditional seed-based community detection algorithm only constructs a single recognition model and cannot effectively identify community members.Aiming at this problem,this paper studies the community detection algorithm based on multi-model ensemble learning and seed set expansion to improve the recognition accuracy of unknown members.The main research contents and conclusions of the thesis are as follows:(1)Community detection algorithm based on important seed member expansion and selfsampling ensemble learningAiming at the problem that the seed node selection in the existing seed-based community detection algorithm is unreasonable and the single algorithm model fails to consider the seed information from many aspects,this paper proposes an algorithm based on the integration of important seed members and self-sampling of bagging.Firstly,the importance of each adjacent node is obtained by calculating the degree centrality coefficient of the node.The unknown members are identified according to the important members and the existing seed set,and then the improved algorithm is used as the base classifier,and bagging self-sampling is used.The algorithm obtains multiple differential base classifiers,and finally combines multiple base classifiers to improve recognition accuracy.The experimental results show that the community detection algorithm with node degree centrality has a strong recognition accuracy.The selfsampling ensemble algorithm is compared with the traditional community detection algorithm.The recognition accuracy is better than the traditional seed collection based community detection.The algorithm has a large improvement and is more suitable for the identification of unknown members in sparse communities.(2)Community detection algorithm based on adaptive enhanced ensemble learningAiming at the problem that self-sampling ensemble learning has not considers ownership of node and community.This paper proposes a community detection algorithm based on adaptive enhanced ensemble learning.From the network topology and AdaBoost adaptive algorithm,the community detection based on seed set is regarded as a two-category problem.The change of training data set distribution depends on whether the samples are correctly classified.First,the initial weight of the sample is given,then the weight of the sample of the correct classification is lowered,and the weight of the classification error sample is increased.Finally,the weights of the multiple weak classifiers are combined to obtain a strong classifier.The experimental results show that the ensemble algorithm has a better recognition effect than the traditional seed-based community detection algorithm,and it has better effects when dealing with dense communities.It shows that the adaptive learning ensemble learning community detection algorithm can improve the recognition accuracy and provide users with more valuable data.