Dans cette thèse, nous étudions les techniques de raisonnement utilisant des graphes d'argumentation générés à partir de bases de connaissances inconsistantes exprimées dans le langage des règles existentielles.Les trois principaux résultats sont les suivants. Tout d'abord, nous étudions les propriétés structurelles des graphes obtenus à partir de bases de connaissances exprimées avec des règles existentielles et nous donnons plusieurs indications sur la manière dont leur génération peut être améliorée. Deuxièmement, nous proposons une technique pour générer un graphe d'argumentation où plusieurs arguments peuvent attaquer collectivement, remplaçant ainsi la relation d'attaque binaire classique et montrons expérimentalement les avantages de cette technique. Troisièmement, nous nous intéressons aux approches fondées sur les classements pour le raisonnement en logique et en argumentation. / In this thesis, we investigate reasoning techniques with argumentation graphs generated from inconsistent knowledge bases expressed in the existential rules language. The existential rules language is a subset of first-order logic in which a knowledge base is composed of two layers: a fact layer and an ontology layer.The fact layer consists of factual knowledge (usually stored in relational databases) whereas the ontology layer consists of reasoning rules of deduction and negative constraints.Since the classical query answering approaches fail in the presence of inconsistencies, we chose to work with an conflict-tolerant reasoning approach that is based on building graphs with structured arguments and attacks from the deductions of the underlying logical knowledge base.The three main results are the following. First, we study how argumentation graphs are obtained from knowledge bases expressed in existential rules, the structural properties of such graphs and show several insights as to how their generation can be improved. Second, we propose a framework that generates an argumentation graph with a special feature called sets of attacking arguments instead of the regular binary attack relation and show how it improves upon the current state-of-the-art using an empirical analysis. Third, we interest ourselves to ranking-based approaches in both the context of query answering and argumentation reasoning. In the former, we introduce a framework that is based on ranking maximal consistent subsets of facts (repairs) in order to have a more productive query answering. In the latter, we set up the foundation theory for semantics that rank arguments in argumentation graphs with sets of attacking arguments.