Blended Skill BotsTalk (BSBT)

BSBT数据集是由延世大学人工智能系开发的，包含30万条对话，旨在通过多技能对话训练提升聊天机器人的沟通能力。数据集通过BOTSTALK框架自动生成，结合了来自Convai2、WoW和ED等多个数据源的技能，支持聊天机器人在对话中展示多种技能，如个性、知识和同理心。BSBT数据集不仅支持技能融合，也保持了特定技能的深度，适用于开发能够灵活应对不同用户和情境的聊天机器人。

The BSBT dataset was developed by the Department of Artificial Intelligence at Yonsei University, consisting of 300,000 dialogues. It is designed to enhance the communication capabilities of chatbots through multi-skill dialogue training. The dataset is automatically generated via the BOTSTALK framework, which integrates skills from multiple data sources including Convai2, WoW, and ED, enabling chatbots to demonstrate diverse skills such as personality expression, knowledge application, and empathy. The BSBT dataset not only supports skill fusion but also maintains the depth of individual skills, making it suitable for developing chatbots that can flexibly respond to different users and scenarios.