Diversity of Content Sources in Social Networks

The jury members are

• Johanne COHEN, Directrice de recherche, Laboratoire Interdisciplinaire des Science du Numerique (LISN), Examinatrice
• Angela BONIFATI, Professeure, Université Claude Bernard Lyon 1, Rapporteur & Examinatrice
• Cedric DU MOUZA, Professeur, Conservatoire National des Arts et Metiers (CNAM), Rapporteur & Examinateur
• Oana BALALAU, Chercheur, Institut National de Recherche en Sciences et Technologies du Numérique (INRIA), Examinatrice
• Noha IBRAHIM, Maitresse de Conference, Université Grenoble Alpes, Examinatrice

Résumé

Recent events have shown how important the social Web has become to our daily lives. Indeed, most information no longer propagates through single, centralized, points of information, such as TV news or newspapers. Instead, it propagates through de-centralized means, through the large – sometimes global – social networks enabled by the Web. The decentralized nature of the Web means that information of all sorts can spread instantly and without filtering, leading to both positive effects, such as when crowds spontaneously organize for common good goals, or negative effects, such as the spread of fake information which might be passed on as valid by otherwise well-meaning consumers.

How can fake information pass on as valid? This can occur when the consumers of information think they have full knowledge due to their social neighbourhood and the information that is disseminated is representative of the world at large. The concept is well-known in the literature: we talk of filter bubbles or echo chambers, where consumers of information are stuck in an environment that reinforces their own point-of-view, but does not challenge it. Compounding this problem are commercial recommendation algorithms that essentially focus on items one has already consumed or which are consumed by the immediate social neighbourhood; as such they do not focus on recommending truly diverse items from diverse sources. This is more evident when the social network is broken in several communities organized around similar interests, or even ideology-based communities that do not communicate with each other.

In this project, we will take the first steps in enabling diverse recommendations in online social networks. We will first focus on recommending content providers to users of a social network. To take an example of a well-known Web application where this is relevant, consider Twitter. It can arguably be thought of as a sort of public square of the Web; indeed, many events have started on Twitter or like applications (consider e.g., the Arab spring); in some cases, the news broke faster that on traditional media channels. Generally, the Twitter feed of a user of the application consists of the tweets of the users that they followed, or the retweets thereof. In a sense, each Twitter user can be considered either a content creator or a content transmitter (via re-tweets). The social network is hence composed of nodes (users, which are both consumers or creators of content), brought together by links (the following/follower relationships).

Then, the problem becomes rather intuitive. We are given a social graph, and the objective is to add links to it, such that the diversity of propagated information is maximized. Our task is to recommend links between users and content sources such that the graph is split in as few communities as possible. This, for a given social network user, means that they will be exposed to as many other communities as possible.