Gutenberg Poem Dataset

Gutenberg Poem Dataset 用于下一个诗歌预测组件。有越来越多的工作在分析和减轻语言理解、生成和检索任务中的社会偏见，尽管在创造性任务中检查偏见的研究仍然不足。创意语言应用程序旨在与用户直接交互，因此量化和减轻这些应用程序中的社会偏见非常重要。我们介绍了一项关于在诗歌创作系统中检索下一节经文建议时减轻社会偏见的管道的新研究。我们的结果表明，通过情感风格转移来增强数据有可能减轻社会偏见。

Gutenberg Poem Dataset is intended for next-poem prediction components. There is a growing body of work analyzing and mitigating social biases in language understanding, generation, and retrieval tasks, yet research examining biases in creative tasks remains insufficient. Creative language applications are designed to interact directly with users, so it is critical to quantify and mitigate social biases in these applications. We present novel research on a pipeline for mitigating social biases when retrieving suggested next stanzas in poetry generation systems. Our results demonstrate that augmenting data via affective style transfer has the potential to mitigate social biases.

搜集汇总

背景与挑战

背景概述

Gutenberg Poem Dataset是一个用于诗歌预测组件的数据集，旨在分析和减轻诗歌创作系统中的社会偏见。它通过情感风格转移增强数据，以帮助减少创意语言应用中的偏见问题。