The Design of an Online Social Network and its Knowledge-Based Recommendation System

Résumé

Over the last decade, Online Social Networks (OSNs) have been growing quickly to become some of the largest systems available. Their users are sharing more and more content, and in turn have access to vast amounts of information from and about others. This increases the risk of information overload for every user.

Information overload usually refers to the difficulty to make decisions when there is too much information. In such situations, the decision-maker is confronted to so much information that it may become unclear what the exact decision problem is, what the alternatives are, how to compare them in order to single one out.

We are interested in how to mitigate information overload for users of OSNs. Specifically, we are interested in how to design an OSN and its Recommendation Systems (RSs), which filter out and prioritize content for users, and thereby take over some of the decision-making effort from the user. The overall idea is that there can be an overload of information available to users of an OSN, and we want to design an OSN with an integrated RS which processes the information, so as to recommend to the user only those which, according to some specific criteria are the most relevant to that user.

In order to address this general question, we focus on the Requirements Elicitation, Requirements Modelling, and the Analysis and Modelling of the Dynamics observed in OSNs.

More specifically, the requirements for an OSN and its RS are identified in Part II by analyzing popular OSNs and identifying their recurrent features. Bachelor students of the University of Namur were surveyed in order to prioritize these features. Drawing on these results, rules for relevant recommendations are proposed. Part III focuses on the modelling of the elicited requirements, using i*. We model the features, the rules and identify the limitations of i* for the modelling of the dynamics. We propose an extension to i*, allowing for the modelling of this dynamics. We also propose the Content Recommendation Game, giving another perspective to the content recommendation issue.

Given the nature of the samples for the exploratory study, namely Bachelor students of the University of Namur, not all the results are generalizable. But this work is the first step in addressing the general question: "How to design an OSN and its RS which decides in place of the user, what is more or less relevant for her, and consequently filter out some of the potential content that the OSN could deliver?"

la date de réponse	21 avr. 2017
langue originale	Anglais
L'institution diplômante	Universite de Namur
Superviseur	Stephane Faulkner (Promoteur), Ivan Jureta (Promoteur), Annick Castiaux (Président), Haris Mouratidis (Jury), Anna Perini (Jury), Manuel Kolp (Jury) & Michael Petit (Jury)