Resumen de Systematic review of training methods for conversational systems: the potential of datasets validated with user experience
Carolina Abrantes, Juliana Camargo, José Nunes, Maria João Antunes, Óscar Mealha, Luís Nóbrega
The maturation of artificial intelligence technologies, such as Machine Learning algorithms, Natural Language Processing (NLP), Automatic Speech Recognition (ASR), and Natural Language Generation, is transforming the way users interact with technology. With the increasing prevalence of voice interactions, it is crucial to understand the process of training conversational agents. This systematic review examines how human data is collected for training these agents, with a specific focus on datasets obtained directly from human participation in real-life contexts of need and use. The study follows the PRISMA guidelines, and searches were conducted in Scopus, Web of Science, and ProQuest databases in English, covering the period from 2005 to 2020, to provide a comprehensive overview of published practices until July 2020. A total of 22 papers were included in this review from the search iterations. The primary findings of these papers indicate a common use of learning from demonstration/observation and crowdsourcingmethods in system training and dataset cataloguing. Additionally, techniques such as handwriting and sentence labelling, as well as Wizard-of-Oz based studies, were employed in the research.
