Research On Social Network Entity Link Prediction Technology Based On Graph Embedding

In nowadays world,social network or social media plays a crucial role in one's daily life,and social network have accumulated huge volume of valuable data.Thus how to employ these materials becomes a noteworthy thing.In social networks,users are deemed as nodes,and the interaction between users constitutes the edge between nodes,and thus form a complete network.Besides,in addition to the network formed by users,each user has rich information,such as user profile,location information,text,etc.,to assist researchers to analyze the whole network and individuals in the the network.In this paper,how to use these two kinds of information—topological structure information and non topological information—to analyze social network is studied:First of all,for the exploration of network topology information,the widely used method is network embedding.The basic idea of this method is to map the nodes(or edges)into vectors,which can reflect the topological relationship of the network.The embedding vectors generated by high-quality embedding methods can achieve better results for many downstream tasks such as link prediction,community detection,recommendation,etc.In this paper,the method combining high-order similarity and convolutional self encoder is used to generate embedded vectors,and then we conducted a series of experiment to verify the performance of our model.Result indicate that even though there is variation existing in different kind of datasets,our model can still acquire a stable and excellent performance,especially in large-scale data sets,which surpasses many current strong embedding models.Secondly,for the link prediction task in the network,most studies focus on how to predict the existence of edges or nodes in a complete network.However,when the relationship between the node to be predicted and the current network is completely unknown,that is,there are isolated nodes potentially belonging to the current network,how to explore the relationship between the isolated node and the nodes in the network becomes a problem worth of studying.At the same time,the performance of the model using supervised algorithm in link prediction model is often limited by the high cost of manual annotation,so how to use a small amount of tag information to complete the link prediction task is also a valuable research direction.Based on the above considerations,this paper will combine the topological information and non-topological information of the network,and map the non topological information of the isolated node into the topological information,and then use the semi-supervised model based on the generative adversarial network to realize the link prediction task of the isolated node.The performance of our model on social network data sets proves that our model has a good prediction effect.