Graduation Year

2020

Document Type

Thesis

Degree

M.S.Cp.

Degree Name

MS in Computer Engineering (M.S.C.P.)

Degree Granting Department

Computer Science and Engineering

Major Professor

Adriana Iamnitchi, Ph.D.

Committee Member

Giovanni Luca Ciampaglia, Ph.D.

Committee Member

Lawrence Hall, Ph.D.

Keywords

Generative Adversarial Networks, Graph Anonymization, Graph ConvolutionalNetworks, Graph Learning, User Activity Forecasting

Abstract

Recent advances in neural network-based machine learning algorithms promise a rev-olution in prediction tasks across a variety of domains. Of these, forecasting user activity insocial media is particularly relevant for problems such as modeling and predicting informa-tion diffusion and designing intervention techniques to mitigate disinformation campaigns.Another potential task is anonymizing social network datasets to facilitate their distributionand promote research. Given the success of deep generative models, it may be possible touse them for anonymization. Social media seems an ideal context for applying neural net-work techniques, as they provide large data sets and challenging prediction objectives. Yet,our experiments find a number of limitations in the power of deep neural networks and tra-ditional machine learning approaches in predicting user activity on social media platformsas well as creating anonymized networks. Two studies are conducted in this work. Thefirst describes whether a Generative Adversarial Network could produce slightly dissimilarattributed graphs from an original which may implicitly anonymize it. We find issues inhow the graph is assembled and how the generator learns attributes for nodes. The secondstudy describes the challenges we encountered while attempting to forecast user activity ontwo popular social interaction sites: Twitter and GitHub. The custom sequence-to-sequencearchitecture that is used suffers limitations related to dataset characteristics, specificallytemporal aspects of user behavior.

Download

Share

COinS