Conversational Interactive Recommendation

• Autor(a) principal: Correia, Alexandre Teixeira Lobato
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/151149
• topic: Conversational Assistant; Conversational Agent; Task-Oriented Conversational Agent; Risk-aware Conversational Agent; Reinforcement Learning; Deep Reinforcement Learning; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

When we search for something online, sometimes we know what we want, but we do not know how to exactly describe it. To address this challenge, this thesis presents a framework based on reinforcement learning, which improves conversational assistants in order to get closer to the user’s desired product by asking a sequence of clarifying questions and giving recommendations. These clarifying questions are chosen before each interaction, in real-time, and are capable of narrowing down the search space, helping the user finding what he wants. In order for the assistant to decide which clarifying question to use, its neural network needs some information about the recommendation conversation. After the selection of the question, the assistant generates the question using a template, which might need some recommended products. Having generated the question, the assistant presents it to the user and gets his answer, from which extracts some information to update the recommendation conversation history. This framework can be served as a module inside projects and lets the recommendation conversation to evolve in various directions: minimizing the number of interactions, minimizing the repetitiveness of used questions and minimizing the repetitiveness of used questions according to their feedback times.