Research On Dynamic Social Network Alignment Based On Representation Learning

With the development of the Internet,more and more people are willing to make friends and live in virtual social network.In order to enjoy different functions of social networks,many users join different types of social network platforms,which forms a phenomenon that a user owns multiple social network accounts.However,the closure of different platforms causes the data generated by the same user to be disconnected,which brings difficulties to the data mining tasks across social networks.Therefore,user alignment across social networks has become a key issue to be solved.Its main goal is to find out the accounts belonging to the same person in different social networks,so as to fuse the data in different platforms,and further provide basic support for social network applications such as user recommendation and community detection.Network alignment tasks are mostly based on the static network environment,ignoring the inherent property of network dynamic change.Dynamicity is one of the core attributes of social networks.Modeling dynamic networks can better reflect network changes and reduce obsolete data.However,most models need to be retrained when updated,which causes the waste of time and resources.Based on this,this thesis studies how to model user alignment in dynamic network environment.The main contents and innovations of this thesis are as follows:1.A shallow dynamic social network alignment model based on heuristic algorithm is proposed.Firstly,the attention mechanism is used to obtain the local importance weight of the new node in a single network.Secondly,the anchor nodes are adopted as supervised information for heuristically learning the alignment task driven local influence of new nodes.Finally,by preserving the second-order similarity of the network,the model incorporates the learned weight into the local network updating process to achieve the goal of dynamic user alignment across networks.Experimental results on real data sets show that the proposed model has as good performance but lower time complexity compared with several state-of-the-art algorithms.2.A deep dynamic social network alignment model based on GCN(Graph Convolution Networks)is proposed.Firstly,a fusion network is formed by the structural relationship between nodes and edges in the source network and the target network.According to the time series of dynamic network,a series of network snapshots are generated.Secondly,the adjacency matrix of the network is input into the GCN layer and GRU(Gated Recurrent Unit)layer of the deep neural network.Finally,a loss function is defined in the full connection layer to classify the nodes,and the potential anchor nodes and non-anchor nodes are separated.Experimental results on real data sets show that this model can effectively save multi-dimensional information such as network structure information,attribute information and time information,and solve the problems of retraining of dynamic network user alignment task model and single training information.