[cs-talks] Proposal Defense Announcement - Mehrnoosh Sameki

Devits, Christopher R cdevits at bu.edu
Fri Mar 31 12:22:42 EDT 2017

PhD Thesis Proposal Defense
Candidate: Mehrnoosh Sameki

Date: April 3, 2017
Time: 10am
Location: MCS 148, 111 Cummington Mall, 1st floor

Title: Analysis of Political Discourse in Image, Video, and Text via Crowdsourcing

Abstract: Based on a report by BBC News Agency, fake stories like "Pope Francis Shocks World, Endorses Donald Trump for President" attracted millions of interactions (e.g., views/shares/likes), and trended as top stories on Facebook social media. Facebook went under fire over fake news and was accused of “prioritizing the popular topics and not knowing how to distinguish between real and unreal stories”. In this proposal, we consider the task of fact checking of public figures' claims, and propose to use crowdsourcing to uncover the truthfulness of a claim. To achieve this goal, we train a machine learning system to identify the influence of opinion bias of crowd workers on their perceived reality of a claim content by predicting each worker's projected annotation accuracy. We then use the predicted annotation accuracy metric to correct for such opinion bias in produced annotations and obtain the truthfulness of a claim.

We next study the third US presidential debate and investigate how Twitter users react to candidates verbal and nonverbal attributes in real time. We use crowdsourcing to collect a detailed coding of the third US presidential debate between Hillary Clinton and Donald Trump. We sync three datasets: (1) a shot-by-shot nonverbal elements of both candidates during the debate (e.g., body gestures and facial expressions), (2) the volume and sentiment of the corresponding real-time tweets, synced and lagged. and (3) the verbal elements of presidential debates transcripts (e.g., topics of discussion, candidates' attributes) to identify the influence of the candidates’ verbal and nonverbal features on viewers’ tweet content. In order to study the sentiment of real-time political tweets, we propose a budget-efficient crowdsourcing platform that uses machine learning to perform a dynamic allocation of number of crowd workers employed.


Margrit Betke
Computer Science, Boston University

Stan Sclaroff
Computer Science, Boston University

Danna Gurari
School of Information, University of Texas at Austin

Prakash Ishwar
Electrical and Computer Engineering, Boston University

Lei Guo
College of Communication, Boston University
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://cs-mailman.bu.edu/pipermail/cs-talks/attachments/20170331/f2862e65/attachment.html>

More information about the cs-talks mailing list