Conversa LSD - Leveraging Trust and Distrust for Sybil-Tolerant Voting in Online Social Media

Caros,

Na Conversa LSD desta semana (quarta, 16h, no auditório do LSD) temos uma palestra de um visitante, e gostaríamos de convidar a todos. Segue abaixo uma breve apresentação do palestrante e da palestra.

[]s
Nazareno

Nitin Chiluka está no último ano de seu doutorado na TUDelft e trabalha no uso de análise de grafos para entender e projetar sistemas distribuídos, de recomendação e mídia social online. Eu participo da orientação de Nitin e ele está visitando a UFCG no contexto de um projeto CAPES/Nuffic entre TUDelft e UFCG. 

O resumo da apresentação:

Leveraging Trust and Distrust for Sybil-Tolerant Voting in Online Social Media

Voting is a vital component of online social media (OSM).  Votes on content items in OSM, e.g., likes in YouTube and Facebook, favorites in Flickr, and diggs in Digg) are typically incorporated into many of their central features such as recommendations, ‘most popular’-like pages and ranking search results. Voting helps in determining popularity and trustworthiness of content. 

At the same time, due to their open membership access, voting on content items in OSM is susceptible to Sybil attacks.  Malicious attackers can create multiple Sybil identities to outvote the real users of the system. To defend against such an attack, we leverage (i) trust which is inherent in the social network among users in OSM, and (ii) distrust between honest users, who identify some of the spam content items, and the Sybil identities who promoted them. Modeling trust and distrust in the system as a signed network, our method proceeds in two phases. First, we identify nodes and edges that constrain paths along positive edges between the endpoints of each negative edge. Second, we limit the votes from Sybil voters whose paths to honest nodes pass across these bottlenecks. Our simulation results on popular OSM datasets show both the feasibility of incorporating distrust alongside trust to defend against Sybil attacks, and that our method outperforms the state-of-the-art approach, SumUp.