Concepts Approximation Through Dialogue with User

摘要

This paper is an attempt to show how dialogue between a system and a user is important to design a robust system which can learn a user's perspective and revise its knowledge base through interactions. The dialogue is crucial in order to better respond to a given query of a user. The problems, where dialogues can play a role, are discussed from two aspects. One is the aspect in which the system becomes able to learn the perspectives of the user(s) and improve its quality of classifications. The other is the aspect where the system can help a user to get answers to its queries. We have, in particular, considered the problems of (i) learning a user's ontology of concepts, (ii) explaining the system's own classification for a cluster to the user in order to get feedback, and (iii) generating a global description for a cluster, in a user-friendly language, based on a sample of objects available to the system.