shanya/crd3
收藏资源简介:
CRD3数据集,又称关键角色龙与地下城数据集,包含了159集Critical Role节目的转录对话,总计398,682个对话回合。该数据集还包含了来自Fandom wiki的相应摘要。数据集语言独特,完全通过玩家合作和口头互动生成叙述。每个对话包含大量回合、多个抽象摘要以及与先前对话的语义联系。该数据集适用于摘要、文本生成和填充掩码等任务,特别适用于对话建模。
CRD3 dataset, also referred to as the Critical Role Dungeons & Dragons Dataset, contains transcribed dialogues from 159 episodes of the Critical Role show, with a total of 398,682 dialogue turns. This dataset also includes corresponding summaries sourced from the Fandom wiki. The dataset features a unique linguistic style, where narratives are entirely generated through collaborative player engagement and verbal interactions. Each dialogue entry includes numerous conversational turns, multiple abstract summaries, and semantic links to prior conversations. This dataset is applicable to tasks such as text summarization, text generation, and mask filling, and is particularly suitable for dialogue modeling.
数据集概述
数据集名称
- pretty_name: CRD3 (Critical Role Dungeons and Dragons Dataset)
语言和许可证
- language: en
- license: cc-by-sa-4.0
数据集来源和任务
- source_datasets: original
- task_categories: summarization, text-generation, fill-mask
- task_ids: dialogue-modeling
数据集大小
- size_categories: 10K<n<100K
数据集描述
数据集总结
- 内容: 包含159集Critical Role节目的文本对话,总计398,682个对话轮次,以及从Fandom wiki收集的相应摘要。
- 特点: 通过玩家协作和口头互动生成的叙事,每个对话包含大量轮次和多个摘要。
支持的任务和评估
- 任务: 抽象摘要
- 评估: 使用fast abstractive summarization-RL模型,ROUGE-L-F1得分为25.18。
语言
- 语言: 英语
数据集结构
数据实例
- 示例: 包含对话轮次、发言者名称和发言内容等。
数据字段
- 字段: chunk, chunk_id, turn_start, turn_end, alignment_score, turn_num, turns (包含names和utterances)
数据分割
- 分割: train (26,232), validation (3,470), test (4,541)
数据集创建
数据收集和标准化
- 来源: 从Critical Role节目收集,包含159集的转录文本。
- 摘要来源: 从Critical Role Fandom wiki收集。
语言生产者
- 生产者: Critical Role节目的演员。
许可证和引用信息
许可证
- 许可证: Creative Commons Attribution-ShareAlike 4.0 International License
引用信息
- 引用格式:
@inproceedings{ title = {Storytelling with Dialogue: A Critical Role Dungeons and Dragons Dataset}, author = {Rameshkumar, Revanth and Bailey, Peter}, year = {2020}, publisher = {Association for Computational Linguistics}, conference = {ACL} }




