Dungeons and Dragons Play-By-Post Dataset

该数据集由近900场《龙与地下城》游戏创建而成，包含了800,000个对话轮次、500,000次骰子投掷以及5800万字的文本，并标注了部分游戏状态信息。数据集不仅包含了角色内对话和角色外对话，还通过启发式方法提取了游戏状态特征，以便用于训练机器学习模型。该任务旨在进行《龙与地下城》中的对话生成和游戏状态预测。

This dataset is constructed from nearly 900 Dungeons & Dragons (D&D) tabletop role-playing game sessions, encompassing 800,000 dialogue turns, 500,000 dice rolls, and 58 million words of text, with partial game state information annotated. It includes both in-character and out-of-character dialogues, and extracts game state features via heuristic methods for training machine learning models. The corresponding task targets dialogue generation and game state prediction in Dungeons & Dragons scenarios.