Title: Understanding Social Media Credibility

Tanushree Mitra

Ph.D. Student
School of Interactive Computing
College of Computing
Georgia Institute of Technology

Date: Thursday, April 2, 2015

Time: 2:00 PM - 4:00 PM EDT

Location: TSRB 223

Committee:

Dr. Eric Gilbert (Advisor), School of Interactive Computing, Georgia Institute of Technology

Dr. Amy Bruckman, School of Interactive Computing, Georgia Institute of Technology

Dr. Jacob Eisenstein, School of Interactive Computing, Georgia Institute of Technology

Dr. Scott Counts, Nexus Group, Microsoft Research

Abstract:

Online social networks provide a rich substrate for disseminating information through social ties. They have the unique ability to nurture looser but extensive social ties and can foster shorter path lengths for information flow. Together this has led to a scenario where individuals are increasingly relying on online social networks to share diverse information quickly, without relying on established official sources. This ease of information flow while on one hand empower us with unparalleled information access, on the other hand presents a new challenge — the challenge of ensuring that the unfiltered information originating from unofficial sources is credible. The goal of my work is to improve the understanding of credibility of social media information by identifying social media signals that can serve as useful credibility markers.

There have been several organized attempts to study this phenomenon, primarily branching into: 1) studying specific events; or 2) extensive quantitative analysis of social media traces corresponding to historically reported cases of rumors. While valuable, these approaches select on the dependent variable – presuming a priori what rumors look like. To address this gap, in this dissertation proposal, I present an iterative framework for systematically tracking the credibility of social media information. My framework combines machine and human computation in an efficient way to track both less well-known and widespread instances of newsworthy content in real-time, followed by crowd-sourcing credibility assessments. Having run the framework for several months on the popular social networking site – Twitter, I have generated a corpus (CREDBANK) of newsworthy topics, their associated tweets and corresponding credibility scores. With an iterative framework and a unique large-scale social media credibility corpus in place, my goal is to design a prediction algorithm (CRED-RULER) that can reasonably assess the credibility of an event. The social media cues captured in my corpus will form the building blocks of my computational model. A proof of concept evaluation performed on this corpus shows the feasibility of this model.