ORACLE: An Ontology Reasoner for Affective Conversation in Long-term Engagement
Summary
When social robots interact with the same user for longer periods of time, memory can increase the personalization of that robot. In this research, episodic memory in long term interaction for social robots is investigated. Theories on human memory and question asking were combined with those from the field of social artificial agents, which resulted in an ontology where events from the user’s life can be encoded in a 5W1H (who/what/where/when/why/how) structure. This allows a dialog manager to form personalized open questions on properties missing in that encoded structure. Furthermore, the model allows inferring information missing from properties – or find contradicting information in properties – on similar episodes, which can be used to further personalize questions. Lastly, concepts of affect, such as sentiment, emotion, and mood, were added so the affective state of the user can be taken into account during question formulation.