Font Size: a A A

Alignment Of Multiple Literature Networks And Link Prediction Across Networks

Posted on:2020-12-21Degree:MasterType:Thesis
Country:ChinaCandidate:L ChenFull Text:PDF
GTID:2480306452472834Subject:Computer technology
Abstract/Summary:PDF Full Text Request
With the popularity of preprint network in the academia,the research on link prediction in preprint network,such as collaborator recommendation and venue recommendation,meeting the needs of practical application.However,the preprint network has much less information than the traditional literature network such as DBLP,which will lead to an unsatisfactory link prediction result.Since the preprint network shares some entities with other literature networks,if the information of other networks can be obtained through these shared entities,it will help to reduce the impact of information sparsity and improve the accuracy of link prediction.In this paper,the link prediction across aligned literature networks problem is proposed as the main research objective,the ar Xiv preprint network and DBLP network are taken as research objects,and puts forward three stage analysis method containing the link prediction in a single literature network,the anchor link prediction across multiple literature networks,and the link prediction across literature networks.In addition,a corresponding solution is proposed for each type of relationship prediction.Finally,a time-aware link prediction across literature networks model in the aligned literature network is completed by combining the time factor.Firstly,in order to predict the link between nodes in a single literature network,this paper aims at the core problem of how to measure the similarity between nodes effectively in heterogeneous literature network,taking the basic idea of take node similarity as the probability of relationship between node,using the network representation learning method to embed the nodes in the literature network into a low-dimensional space and compute the similarity between nodes,and propose a network representation learning model based on meta structure,named as MSNRL.Secondly,in order to predict the anchor link across multiple literature networks,HANAM model based on hashing is proposed.HANAM combines structural features and attribute features to represent nodes,adopting the locality sensitive hashing hash to constructs the sparse weighted bipartite graph of the anchor link prediction,and then the redundant links are pruned according to the boundary constraint,balance constraint and maximum similarity optimization function,the anchor link results can be obtained.In this way,multiple literature networks are linked together to provide an infrastructure for link prediction across literature networks.Finally,in order to solve the problem of link prediction across literature networks,the aligned network is used as the infrastructure of the link prediction across networks,taking the network with weak information completeness as the target network and the network with high information completeness as the source network,the model of link prediction across networks from the nodes of target network to the nodes of source network is realized by combining the random walk strategy and the information transfer about the similarity between nodes for anchor nodes.Further,taking into account the time factor,this paper introduced the time-aware matrix,then measure the relationship between nodes across literature networks in combination with the topology information of networks,the semantic information of nodes and time factor.A series of experiments show that,the method in this paper combining the topology information of networks,the semantic information of nodes and time factor in the heterogeneous literature network based on the meta structure,the random walk strategy and the time-aware matrix,it can not only realize the link prediction in s single literature network with high quality,but also realize the link prediction across literature networks by anchor node alignment.
Keywords/Search Tags:aligned networks, link prediction, network representation learning, random walk
PDF Full Text Request
Related items