Towards a Persona-coherent Conversational Agent

22/11/2021

Dialog systems have been at the center of Natural Language Processing (NLP) since its inception. With a wide range of applications, this type of system is particularly interesting as a user interface, creating the possibility of a more natural and convenient user experience. In the context of Customer Support,goal-oriented dialog systems are now widely used, helping users carry out specific tasks. Traditionally, these systems were created by employing knowledge-based architectures. However, the growth of deep learning and the increase in data availability facilitated the development of neural dialog systems, which can be trained end-to-end. A well-known example of such a system is the “Transformer”, a self-attentional model that has achieved state-of-the-art results in multiple NLP tasks. Notwithstanding, these systems still present some shortcomings, particularly in terms of scalability. The need for large amounts of data and considerable computing power can be an impediment, especially in situations where multiple entities must be represented. In Goal-Oriented Dialog Systems, this becomes evident when considering multi-brand Customer Support, since each brand must communicate differently with its users, meaning one model must be developed and maintained for each brand. In Open-Domain System, an analogous problem arises when considering settings where multiple characters must be impersonated.

Autores da comunidade :

• 

  G

  Gonçalo Nuno Ventura de Melo

  ist187660

Orientadores desta instituição:

• 

  M

  Maria Luísa Torres Ribeiro Marques da Silva Coheur

  ist24630

• 

  A

  André Filipe Torres Martins

  ist46911

electrical-engineering-electronic-engineering-information-engineering - Engenharia Eletrotécnica, Eletrónica e Informática

eng - Inglês

Instituto Superior Técnico