Skip to content Skip to navigation
Qualifying Exam
5/28/2019 03:00 pm
CoRE B (305)

Representation Learning for Predicting Information Spread in Dynamic Heterogeneous Networks

Vahid Azizi, Dept. of Computer Science

Examination Committee: Prof. Mubbasir Kapadia (Chair), Prof. Gerald De Melo, Prof. Mridul Aanjaneya, Prof. Aaron Bernstein

Abstract

Online Social Networks (OSN) like Twitter, Reddit, Facebook, Instagram and the like have become a hotbed for sharing various types of information such as text, image, video and audio. Social interactions on a day-to-day basis change the rate and style of information spread. The accurate and scalable simulation of the spread and evolution of online information provides us a quantitative analysis over both within and across information environments. Particularly, identifying viral content, influential users, misinformation campaigns, fake news etc. in OSN is of significant importance to users, social media sites like Twitter and governments alike. Current Research has targeted similar tasks using efficient graph representation algorithms. However, focus is highly on static graphs where relations do not change with time and neither does the network structure. Also, most of these approaches constraint the network to be homogeneous, where there is a single type of node connected by a single type of edge. These assumptions misrepresent real world networks that are mostly dynamic and heterogeneous. We are investigating representation learning techniques to model the evolution of information spread within and across dynamic and heterogeneous networks. In this talk, techniques used to solve the aforementioned problems with preliminary results obtained as part of the DARPA SocialSim project will be discussed. Limitations of current research and future directions as part of my thesis will also be outlined.