Information network representation learning and its applications

Abstract

Information networks such as social network, citation network, email communications network, etc are becoming more and more popular and attracts a lot of research because their applications on real-life. Representation learning attempts to build a low-dimensional representation for each object in their information network but still preserving the semantic information. In this thesis, we're going to apply deep learning model to learn the representation and solve the problems in information networks analysis such as link prediction, community detection.

In particular, the main contributions of the dissertation proposal are as follows. - We study the information network and review state-of-the-art methods in learning the representation of such networks. Besides, we also analyze and evaluate the advantages and disadvantages of each representation learning technique - We apply representation learning techniques to solve network alignment/anchor link prediction problem in information networks. Specifically, we have propose three models to handle the network alignment problem using a combination of multiple different network representation techniques. - We have experimented our model in real-life data sets. We also evaluate and compare experimental results with the baseline models. The experimental results show that our models give better results than the baseline models.

Information networks such as social network, citation network, email communications network, etc are becoming more and more popular and attracts a lot of research because their applications on real-life. Representation learning attempts to build a low-dimensional representation for each object in their information network but still preserving the semantic information. In this thesis, we're going to apply deep learning model to learn the representation and solve the problems in information networks analysis such as link prediction, community detection.

In particular, the main contributions of the dissertation proposal are as follows. - We study the information network and review state-of-the-art methods in learning the representation of such networks. Besides, we also analyze and evaluate the advantages and disadvantages of each representation learning technique - We apply representation learning techniques to solve network alignment/anchor link prediction problem in information networks. Specifically, we have propose three models to handle the network alignment problem using a combination of multiple different network representation techniques. - We have experimented our model in real-life data sets. We also evaluate and compare experimental results with the baseline models. The experimental results show that our models give better results than the baseline models.