Linguagem MoLIC para o design de agentes conversacionais: aplicabilidade e extensão

• Ano de defesa: 2023
• Autor(a) principal: Ulisses da Silva Fernandes
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ICX - DEPARTAMENTO DE CIÊNCIA DA COMPUTAÇÃO; Programa de Pós-Graduação em Ciência da Computação; UFMG
• Palavras-chave em Português: Agentes Conversacionais; Chatbots; Modelagem de Agentes Conversacionais; Computação – Teses; Chatbot – Teses; Chatbot – Modelagem– Teses
• Link de acesso: http://hdl.handle.net/1843/62174; https://orcid.org/0000-0001-5271-8645

Intelligent conversational agents are becoming more and more popular in different usage domains. The use of this type of technology offers challenges and opportunities for Human- Computer Interaction (HCI), since it is based on conversations in natural language and can present different degrees of intelligence and autonomy. One of these challenges is to investigate whether existing design techniques are suitable for modeling user-agent interaction. In particular, in this research we focus on MoLIC (Modeling Language for Interaction as Conversation), a design-phase dialogue model based on Semiotic Engineering theory, which allows designers to represent interaction as conversations between a system and its users. To conduct our investigation, we initially carried out two case studies in which we generated the MoLIC interaction diagram by reverse engineering two conversational agents – the ANA chatbot, created to support triage and provide information about COVID-19; and Bixby, a virtual assistant for Samsung Electronics smartphones. The purpose of our reverse engineering analysis was to examine and how the interactive aspects of these conversational agents could be expressed in MoLIC. The results of these two studies showed that, although MoLIC was able to express the general interaction model, there was some limitations – related to the expressiveness of the language or its inadequacy to represent these types of system, resulting in a possible overloaded model that could hinder the interpretation and reduce the epistemic nature of MoLIC. Therefore we proposed some adaptations and the creation of new elements in order to adapt MoLIC to this technology and we evaluated these changes by applying them to a TSE chatbot for the 2022 Brazil presidential elections and in an online shopping website (Lojas Americanas) to verify if the changes were able to model not only conversational agents, but also other systems kinds. The contributions of this study focus on (1) identifying the applicability of MoLIC to model this type of technology and pondering considerations on how to extend or adapt MoLIC to overcome them; and (2) direct the HCI community to issues and new initiatives that can help designers design and model intelligent conversational agents.