Research And Application Of Active Semi-supervised Network Structure Exploring Algorithm

In the information age,data from social platforms is generated continuously every day,and many data can be modeled as large-scale networks.Based on these network data identifies the potential multi-type clustering structure quickly and accurately,which is beneficial to practical applications.Such as the recommendation system can use the clustering information of users for recommendation,and the public opinion analysis can utilize user clustering to discover the user interaction mode.The existing community structure discovery algorithm can only identify the community structure of the network,for multi-type structure of the network,although the unsupervised network structure discovery algorithm can discover its structure,the accuracy is not high.Prior information helps to improve the accuracy of network structure discovery,but prior information is difficult to obtain,and requires a lot of cost.Active learning based on the sampling strategy selects the supervision information that maximizes the promotion of performance of the model,obtains high-quality priors at a lower cost,further improve the accuracy of the network structure discovery algorithm.Therefore,the active semi-supervised network structure discovery algorithm has more research and application value.The unsupervised online EM algorithm onlineVEM can find large-scale network multi-type structures,but it relies heavily on the pros and cons of the initial parameters of the model,especially when the network structure is complex,the results are unstable and inaccurate.Therefore,combining the onlineVEM algorithm with the uncertainty sampling strategy,an active semi-supervised network structure discovery algorithm ASonlineVEM based on iterative framework is proposed.The algorithm initializes the model based on the representative nodes;then iteratively executes three tasks: running the online algorithm onlineVEM,actively selecting nodes,labeling nodes and updating parameters of model until the algorithm reaches the threshold or convergence.Experiments on artificial networks and real networks with different structures show that ASonlineVEM algorithm is better than similar algorithms.ASonlineVEM algorithm improves the accuracy of discovering multi-type clustering structure of network to a certain extent,but it only select the prior according to the uncertainty sampling criterion,this can not maximize the promotion of performance of network structure discovery.In this regard,the onlineVEM algorithm is combined with BALN algorithm based on batch active learning,and an active semi-supervised networkstructure discovery algorithm BAOE is proposed.It include three node importance criterias,uncertainty,representativeness and centrality,to actively select node to label,and the node set is expanded based on the random walk strategy.Experimental results on artificial networks and real networks show that the BAOE algorithm can select the set of nodes that maximize the performance of the algorithm.Finally,in order to further verify the validity and practicability of the BAOE algorithm,the designed algorithm is applied to the CSDN user network structure discovery to identify the potential cluster structure in the user network.we analyze network user characteristics base on the clustering results.The experimental results show that the BAOE algorithm has certain practical application value.