Dear All,
You are cordially invited to the upcoming seminar with details below
Speaker: Alvaro Garcia-Recuero
Title: Privacy vs Abuse detection in Decentralised Online Social Networks
Abstract:
Today supervised learning algorithms are widely used for abuse detection, namely spam, bot and even hate speech in centralised Online Social Networks. However, literature investigating the utility of supervised learning to detect abuse in future Decentralised Online Social Networks is scarce. In a decentralized setting, there is no central authority to control the infrastructure that collects and protects the privacy of users’ metadata; so that it is in the hands of individual participants to employ protocols that protect privacy of metadata correctly. Decentralised sites as Twister already provide a similar user interface to a centralized one as Twitter, but with a very different underlying architectural design to protect such metadata. To start assessing the Privacy vs Abuse trade-off, we collect a corpus of messages and social relations from Twitter to identify which metadata we need to consider sensitive or private in future decentralised settings. In practice, this means reducing the number of features available to learning algorithms for abuse detection. We only collect metadata available in the vicinity of the potential victim, namely at most at one-hop distance in the network. For the metadata that we consider private or sensitive, we design a privacy-preserving protocol that protects it in order to avail its use with the supervised learning algorithms. We show how using supervised learning for abuse detection is compatible with a future Decentralised Online Social Network design, effectively avoiding the sort of Big Data mining performed by large centralised Online Social Network providers. In addition, our latest results apply “data minimisation” techniques that achieve more efficient computation of the privacy-preserving protocol and still yield similar abuse detection rates.
Bio:
Álvaro García-Recuero holds a PhD in Computer Science by University of Rennes 1, developed under the supervision of Christian Grothoff (PhD UCLA) and awarded by an international jury composed of Bogdan Carbunar (Florida International University), Carmela Troncoso (IMDEA Software) and Pierre-Alain Fouque (UR1).
His background is in Distributed Systems and his research interests range from the design of Privacy-Enhancing Technologies for Online Social Networks to the measurement of Big Data frameworks and Energy Efficiency issues in networked systems. Since 2013, holds a double joint Master Degree in Distributed Computing with the oldest and largest engineering school in Portugal (IST) and one of the most renowned in Europe (KTH). Prior to that, he worked in a number of multinationals and high-tech software development companies, obtaining a valuable career track-record while graduating with Upper Merits on a BSc in Information Systems & Information Technology at Dublin Institute of Technology.
He has published research related to Cloud Computing, Online Social Networks and Privacy at ACM/IEEE/Springer recognised venues such as CloudCom, ASONAM and ESORICS Data Privacy Management workshop respectively. He releases his research contributions as Free Open Software on his github, thus trying to support free software and support reproduceable scientific research. He is always interested in bold, challenging ideas that push-forward innovation and research.
For more details, please go to the link
http://networks.eecs.qmul.ac.uk/category/seminar/
Venue: Francis Bancroft 2.41
Time and Date: 2 to 3 PM, Wednesday, 27/09/17.
Any comments, advice would be appreciated.