HonkaiChat: Companions from Anime that feel alive!

TL;DR

The work targets the gap between static, reactive anime-themed chatbots and the dynamic, evolving nature of human conversation by introducing event-driven prompts and character-specific fine-tuning. It trains a Honkai: Star Rail-based bot (March 7th) using curated events and lore-grounded data, leveraging 5,000 targeted training instances on a LLaMA 3.1-8B backbone with LoRA PEFT. Across GPT-4 qualitative evaluations, event-driven prompts improve engagement and reduce hallucinations, suggesting richer, more believable interactions in role-playing contexts. The findings indicate strong potential for dynamic, event-aware agents in gaming and fan communities, with notable avenues for improved alignment, personality diversity, and scalable deployment.

Abstract

Modern conversational agents, including anime-themed chatbots, are frequently reactive and personality-driven but fail to capture the dynamic nature of human interactions. We propose an event-driven dialogue framework to address these limitations by embedding dynamic events in conversation prompts and fine-tuning models on character-specific data. Evaluations on GPT-4 and comparisons with industry-leading baselines demonstrate that event-driven prompts significantly improve conversational engagement and naturalness while reducing hallucinations. This paper explores the application of this approach in creating lifelike chatbot interactions within the context of Honkai: Star Rail, showcasing the potential for dynamic event-based systems to transform role-playing and interactive dialogue.

Figures (4)

• Figure 1: Conversation example starter and reply
• Figure 2: The evaluation result of our model vs base model
• Figure 3: A typical conversation starter for interacting with LLMs
• Figure 4: The scores of conversations for engagement for with and without events