Evaluating a Chatbot-Based Realtime Interactive Visual Novel

Abstract

This paper introduces a realtime interactive visual novel game that enables free-form conversations with NPCs using chatbots and language models. The client-side of the game was developed using the Unity game engine, and a realtime dialogue system with NPCs was implemented using Amazon Lex and the DialoGPT language model. This dialogue system classifies the intent and sentiment of the player's input text, generates appropriate responses based on the identified intent, and adjusts the NPC's affinity level according to the sentiment analysis results. To evaluate the game, a small-sized pilot study was conducted with a total of 16 participants. The experiment results showed that the participants responded most positively to the version of the dialogue system that combined chatbots and language models. Additionally, it was found that players preferred NPCs that responded proactively rather than passively avoiding conversations when failing to understand the dialogue flow.

Keywords: Visual Novel Game, Chatbot, Interactive Narrative