Researches On The Negative Link Prediction In Signed Social Networks And On Network Embedding

In the current Internet era,massive network information data is continuously generated with the popularity of intelligent social media and diverse online applications.How to capture effective information and analyze it intelligently in complex data is a new and lasting research topic in the era of big data and artificial intelligence.The generation of Big data on the network is closely related to the explosive growth of the number of Net users.At the same time,more and more people has changed their living behaviors from being in the real space to cyberspace.The large amount of user data information contained in the signed social networks is worth mining and discovering.The links between users are not only an intuitive reflection of social network structure,but also the flow paths of information data in social network.Therefore,the research on users' links is of positive significance for the development of social network and related online applications.However,negative link prediction in signed social networks is a relatively inadequate research branch in the field.Compared with positive links,negative links still play an important role in social networks.Therefore,this dissertation focuses on the mining of negative links information in signed social networks.In the era of big data,the relationship networks between entities are no longer simple relationships between users.The complex relationships between people,things,people and things have become the essential feature of current network data,and the dimension of network data is also increasing.This poses a new challenge to data mining,and brings uncertainty in the field of data analysis related to artificial intelligence.This dissertation extends the research of social network to general information network.By studying the complex high-dimensional network embedding,the low-dimensional representation of data network is realized.Obviously,with the continuous growth of information data,network embedding has an urgent academic significance and application value for current big data research.Based on data mining in data networks,this dissertation focuses on the negative link prediction in signed social networks and the information networks embedding.The main contents and innovations of this dissertation are as follows:(1)Distrust Prediction in Signed Social NetworkThe rapid growth of social media services brings a large amount of social media data,which contains a lot of valuable information that is worth mining and discovering.In social network,the discovery of user relationship has attracted great attention of researchers.The discovery of user relationships in social networks can not only assist community detecting but also improve the accuracy of recommendation systems.It can also predict the flow of social data and conduct public opinion analysis.However,most studies focus on the prediction of trust,the prediction of distrust can't get the same degree of attention.As an important social concept,distrust prediction has an important significance.Distrust prediction not only can help users avoid fake information but also can discover crisis of trust in social or family.Distrust prediction also present some challenges.For example,distrust relationships are very sparse and negative interaction data is very little.To solve these problems,this dissertation only mines the inducing factors of users' distrust relationship from the network topology,and no longer relies on the interactive data between users.A semisupervised distrust prediction model is proposed based on machine learning algorithms and optimization theory.Finally,the validity of the proposed algorithm is verified on real data.(2)Unsupervised Negative Link Prediction in Signed Social NetworksThere are various types of links among nodes in signed social networks,but these links can be roughly divided into two categories: positive and negative links.The positive and negative links not only correspond to the relationship properties of node pairs,but also reflect the structural characteristics of nodes in social networks,so link prediction is also of great significance for sociological research.there are very few negative interaction data;negative links are much sparser than positive links and social data is often noisy,incomplete and fastevolved.This dissertation intends to address this novel problem by solely leveraging structural information and further proposes a framework based on the Projective Nonnegative Matrix Factorization,so as to incorporate network embedding and user's property embedding into negative link prediction.Empirical experiments on real-world datasets corroborate its effectiveness.(3)NEGAN: Network Embedding with Generative Adversarial NetworksEmbedding high-dimensional network data into low-dimensional data structures has been proved successful in many areas,such as community detection,node classification,link prediction and network visualization.However,how to efficiently represent network data in low-dimensional form has always been a research hotspot.This is not only because of the huge increase of data,resulting in the huge and complex network structure,but also the sparse of links in big data networks,which makes it more difficult to extract network feature information.Because of the good performance of generative adversarial network model in extracting feature information,this dissertation proposes a network embedding framework based on generative adversarial network.In the proposed framework,not only the zero-sum game between generator and discriminator,but also a regulator is added to the generative adversarial network.Through continuous correctness of generator by discriminator and regulator,not only the structural information of high-dimensional network can be embedded in low-dimensional representation,but also the relationship attributes of network nodes can be fused into the low-dimensional representation of each node,thus improving the performance of low-dimensional representation in the application field.As shown by the experiments,the proposed framework is competitive with or superior to the state-of-the-art methods on network embedding tasks.(4)Nesting Generative Adversarial Networks for Network EmbeddingIn most current studies of network embedding,more attention is paid to the structural information of network,but little attention is paid to the attribute data of each node in the network.A good network embedding algorithm can not only reduce the dimension of complex network,but also preserve the data information contained in the original network as much as possible.In this dissertation,a new structure of generative adversarial network is proposed.By combining the generative adversarial networks in a nested and progressive way,the low-dimensional embedding of high-dimensional networks is achieved and the attribute information of nodes themselves is added in the process of embedding.The proposed model is validated by the information network of real-world data,and the low-dimensional representation shows a good performance in different application tasks.In summary,this dissertation uses machine learning,deep learning and projective nonnegative matrix factorization to study on the information network.Although the proposed algorithm models have certain research and application value in some applications,some technologies need further research and improvement in the face of the complex and rapid changes of current data.