Phd Canditate :
Assoc. Professor G. Lekakos
PhD Thesis Abstract
We are in front of an evolution from traditional human-computer interactions to a kind of intense exchange between the human user and new generation of virtual or real systems – Embodied Conversational Agents (ECAs) or Affective Robots – which brings the interaction to another level, the “relation level”. These systems are called “Artificial Companions”, that is to say, systems with which the user wants to build long term relationship.
Artificial Companions are cognitive technical systems which possess a certain level of intelligence and autonomy as well as social skills that allow them to establish and maintain long-term relationships with users. They adapt their functionality to the individual user’s capabilities, preferences, requirements and actual needs; are sensitives to the actual user’s emotional state and disposition; conduct helpful and informative dialogs; presenting their self to the respective user as competent and trustworthy partners.
In this PhD research, I design a dialogue manager model based on Deep Reinforcement Learning, which optimizes the dialogue state tracking to recognize the intentions of the user who talks with the artificial companion system with social role, along with the dialogue history, in order to decide when to provide a timely recommendation as the next system action that will lead to the maximum possible cumulative reward and successful dialogue outcome.