An algorithmic framework for avoiding filter bubbles

Details

Supervisors

Jungers, Raphaël M.

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] en science des données, orientation technologie de l'information

Abstract

Recommendation systems are designed to search for content similar to that which a user already has consumed in order to create a recommendation list likely to be of interest to the user. Recently, it has been argued that such systems not only filter out large amounts of available data but also produce a side-effect referred to as a filter bubble - a virtual environment which contains only contents that confirm users' own beliefs. In this master thesis, we propose a mathematical model of the filter bubble and propose an algorithmic challenge to avoid this phenomenon. Based on existing literature on the current flow centrality, we develop an algorithm which solves this challenge. The novelty of our approach is the ability to find sequences of sets of contents which serve as recommendations that lead the user towards unconsumed categories. In an extensive empirical study we applied our algorithm on RTBF data.