KT-KAIST Natural Interaction Project

Development of task-oriented persona-based dialogue generation combining multi-modal interaction and knowledge modeling

This project aims to develop a persona-based conversational system that is capable of proactive dialog generation, under the theme of “Natural Interaction” defined by KT. With the three sub-goals: (1) end-to-end (E2E) dialogue generation, (2) task-oriented knowledge/memory modeling, and (3) multi-modal conversational interaction, this three-year project builds a preliminary persona-based dialogue system (1st year), develops techniques for knowledge modeling and empathetic dialogue generation (2nd year), and implements a task-oriented, multi-modal interaction dialogues (3rd year). The research focus lies on context-awareness of multi-modal conversation and personalization of memory modeling in order to improve the proactive interaction of the dialogue system by leveraging user’s contextual information and short-/long-term memory.