Collaborative Development and Testing of Task-Oriented Conversational Agents

• Autor(a) principal: Simões, Inês Raquel Leandro
• Data de Publicação: 2023
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/163515
• topic: Conversational Agent; Dialog Management Framework; Collaborative Development; Incremental Implementation; Capture and Replay Tests; User simulation; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

The growing popularity of conversational agents in recent years has resulted in the widespread adoption of this technology by various websites and services, establishing them as a ubiquitous presence in our everyday lives. As a result, it has become imperative that developers have access to tools that facilitate the implementation of these agents. Moreover, the stakes for conversational agent performance continue to rise, driven by advancements in the field and higher expectations, making it so that conversational agents must be capable of evolving to meet these standards and innovations. However, the incremental implementation of agents is not a simple matter, and this difficulty is exacerbated in the context of collaborative development, where teams must work together in a seamless and efficient manner to bring their developments to completion. In order to support these accretions, a conversational agent must be built upon a flexible foundation, that properly supports the integration and modification of modular features to the main solution. Otherwise, developers will face numerous difficulties in their efforts to extend or alter their systems. This dissertation strives to address the challenges encountered by developers of com- plex task-oriented dialog systems, while ensuring the deployment of high-quality agents. It aims to mitigate the complexities inherent to collaborative incremental development, facilitate the integration of new features, and enhance the agent’s ability to provide en- gaging and natural conversations while maintaining controlled behavior. Furthermore, it intends to enable comprehensive automated testing as to detect any disruption to the agent’s expected behavior, or other unforeseen issues. The pursuit of these goals resulted in the creation of a collaborative dialog manager framework and a capture-and-replay user simulation testing tool. These contributions were developed in the context of TWIZ team’s participation in the Alexa Prize TaskBot Challenge 2, and facilitated the seamless development and comprehensive testing of their winning conversational agent.