The Study And Application Of Multiplex Graph Based On Graph Convolutional Neural Network

Graph Convolutional Network(GCN)can efficiently extract feature information from complex graph structure data by convolutional operation,which has attracted the attention of many researchers in recent years.However,in the era of information explosion and the Internet of Things(Io T),most existing algorithms for GCN concentrate on a single-layer graph which can no longer handle all kinds of complex multiplex graphs that we meet in real lives.For example,different social softwares are used in daily lives,and the social networks formed by different social softwares are typical multiplex graphs.Each frame in a video can be abstracted into a topological graph according to a specific need,and the frames are also related to each other in multiplex graphs.To handle these kinds of multiplex graphs,this thesis proposes an algorithm framework based on Graph Convolutional Network,which can deal with multiple network data,and achieves good results in two tasks of multiplex social networks link prediction and group activity recognition and prediction.In the link prediction of multiplex social networks task,with the help of attention mechanism,this thesis digs the structural correlations among different layers of the multiplex networks as well as the network structural information of the target layer to make more precise link predictions.Specifically,this thesis introduces three different attentions,namely the intra-layer nodes distance/degree attention,the intra-layer neighborhood attention,and the inter-layer structural attention,to calculate both the influence among nodes in the same layer and the link correlations in different layers.Compared with other stateof-the-art methods which usually require the information of node attributes or edge types,this algorithm only utilizes the topological information of the networks and thus provides a more general link prediction solution for multiplex networks.This thesis conducts comprehensive experiments on several real-world data sets of different scales.By comparing the state-of-the-art link prediction algorithms,this thesis shows the advantages of this algorithm and the effectiveness of different attentions.Group activity recognition and prediction are a challenging task that has a wide range of applications.This thesis proposes a new group activity recognition and prediction algorithm based on the previous multiplex graph convolutional network algorithm framework by treating each individual in the group as a node and establishing a relational graph for the whole group.After detecting and extracting the trajectories of all individuals,a GCN with an attention mechanism of intra-layer nodes and inter-layer structures is established to analyze group activities.Furthermore,different from other algorithms that can only handle the group activity recognition task,our algorithms can make group activity prediction.We introduce multi-layer perception to model the changes of individual positions to obtain more precise predictions.New group activity labels are made on the widely used Volleyball data set to train and test our new prediction algorithm.Comprehensive experiments on this data set fully demonstrate the advantages of this algorithm in group activity recognition and prediction tasks.