A Character-Based Korean Tourist Dialogue System with Proactive Recommendations and Live2D Embodiment

Abstract

This paper presents a character-based Korean tourist assistant prototype that integrates speech recognition, dialogue management, and multimodal embodiment. The system combines client-side voice capture, server-side transcription, retrieval from the Korea Tourism Organization TourAPI, and lightweight policy mechanisms for response generation. Outputs are delivered through a Live2D avatar that provides spoken responses along with recommendation cards and itinerary suggestions. Beyond technical integration, the prototype demonstrates how domain-specific spoken dialogue systems can leverage open APIs and multimodal presentation to offer practical and engaging support in tourism contexts, with implications for user-centered dialogue design.