Um modelo argumentativo computacional para tomada de decisão consensual baseado na formação de conhecimento comum

• Ano de defesa: 2018
• Autor(a) principal: Possebom, Ayslan Trevizan
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Engenharia Elétrica e Informática Industrial; UTFPR
• Palavras-chave em Português: Processo decisório; Processo decisório em grupo; Processo decisório - Modelos matemáticos; Sistemas de suporte de decisão; Otimização combinatória; Árvores (Teoria dos grafos); Teoria dos grafos; Agentes inteligentes (Software); Sistemas multiagentes; Representação do conhecimento (Teoria da informação); Inteligência computacional; Semântica; Engenharia elétrica; Decision making; Decision-making, group; Decision making - Mathematical models; Decision support systems; Combinatorial optimization; Trees (Graph theory); Graph theory; Intelligent agents (Computer software); Multiagent systems; Knowledge representation (Information theory); Computational intelligence; Semantics; Electric engineering; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/3393

Multiagent System consists of a set of intelligent agents, each one with its own knowledge and goals to be satisfied. Those agents interact with each other exchanging messages and reasoning from the information in the interactions. Argumentation systems allow the agents to interact in a social context by sending and receiving arguments. These arguments represent a particular agent’s point of view, corresponding to a justification or explanation about any information or opinion. Arguments are built with a set of premises that lead to a given conclusion throught a mechanism of logical inference. Argumentation-based models allow agents to exchange arguments among themselves and make collaborative group decisions. This thesis addresses the issue of social decision-making by means of dialogues allowing agents for building common-knowledge. When an argument is sent to the group, the level of consensus on its formulas is identified, as well as its intrinsic strength that represents the extent in which the argument as a whole is known and accepted by the group of agents. In related works, it is observed that there is a lack of deeper analysis of the consensus on the arguments, especially in the case of consent and consensual decision-making. Those works do not either show how to evaluate the acceptability of arguments that are consensually supported. The proposal of this thesis is based mainly on the calculation of the consensus level of the group of agents regarding the information present in an argument (premisse and conclusion). The overall strength of the arguments indicates the extent an argument influences the decision making process. The objective of this thesis is to determine, by means of dialogues among the agents, which arguments best justify a given decision alternative and the partial ordering of these alternatives according to the preference of the group. With this objective, it was proposed a model for consensual decision-making based on argumentation that considers: a set of agents with their respective expertise values; a set of possible decision alternatives; a set of dialogue tables and, through a process that maps the dialogue tables to argumentation graphs, the identification of the overall strength of the arguments and the application of argumentation semantics so that the group’s preferred decision alternative is obtained. Two case studies demonstrate the approximation of the knowledge of the agents and the consensus of the group on the arguments that defend each decision alternative. It also presents a comparative analysis with related works comparing the use of abstract and structured argumentation, weighted arguments, mechanisms of dialogue, evaluation of decision alternatives and semantic results.