Abstract:
With the advent of social networks, some serious concerns naturally arise. Indeed, while
the open nature of these networks constitutes a tremendous opportunity for people to
share knowledge and information freely, some malicious actors may take advantage of
this openness to mislead, or even harm others. The huge numbers of users and the ever
growing volumes of exchanged data call for tools and methods that can help users
know whom -and how much- to trust or distrust. This task has instigated many
research efforts, yet unfortunately most of them fail to satisfy, at once, these must-have
criteria: distrust handling (and prediction), performances, efficiency, and robustness to
network sparsity. This thesis, and the various propositions made herein, aim to provide
some novel and useful algorithms for trust (and distrust) prediction -algorithms that,
besides being simple and intuitive, are up to the forecited challenges
